Relaxed targeting rules help PIWI proteins silence transposons

Abstract

In eukaryotes, small RNA guides, such as small interfering RNAs and microRNAs, direct AGO-clade Argonaute proteins to regulate gene expression and defend the genome against external threats. Only animals make a second clade of Argonaute proteins: PIWI proteins. PIWI proteins use PIWI-interacting RNAs (piRNAs) to repress complementary transposon transcripts^1,2. In theory, transposons could evade silencing through target site mutations that reduce piRNA complementarity. Here we report that, unlike AGO proteins, PIWI proteins efficiently cleave transcripts that are only partially paired to their piRNA guides. Examination of target binding and cleavage by mouse and sponge PIWI proteins revealed that PIWI slicing tolerates mismatches to any target nucleotide, including those flanking the scissile phosphate. Even canonical seed pairing is dispensable for PIWI binding or cleavage, unlike plant and animal AGOs, which require uninterrupted target pairing from the seed to the nucleotides past the scissile bond^3,4. PIWI proteins are therefore better equipped than AGO proteins to target newly acquired or rapidly diverging endogenous transposons without recourse to new small RNA guides. Conversely, the minimum requirements for PIWI slicing are sufficient to avoid inadvertent silencing of host RNAs. Our results demonstrate the biological advantage of PIWI over AGO proteins in defending the genome against transposons and suggest an explanation for why the piRNA pathway was retained in animal evolution.

Fig. 1. PIWI proteins bind sites containing or lacking canonical seed pairing.

a, Small RNA guides direct eukaryotic Argonaute proteins to complementary targets. nt, nucleotide. b, Binding affinities (K_d in pM) of MIWI, MILI, EfPiwi and mouse AGO2 loaded with piRNA-1 for canonical and non-canonical target sites. c, Left, MIWI, MILI, EfPiwi and mouse AGO2 binding affinities for targets contiguously paired from nucleotide g2. Right, relationship between binding energy ΔG⁰ calculated from K_d (mean of three independent trials) and predicted binding energy ΔG⁰. Goodness-of-fit for linear regression (r²) and P value for two-tailed permutation test for Pearson’s correlation are shown. All data are in Supplementary Fig. 2a. d, MIWI, MILI, EfPiwi and mouse AGO2 binding affinities for nine-nucleotide complementary stretches contiguously paired from all guide nucleotides. All data are in Supplementary Fig. 2b. Mean and standard deviation of data from three independent trials are shown (b,c (left), d).

Fig. 2. PIWI slicing tolerates mismatches with any target nucleotide.

a, Change in pre-steady-state cleavage rate for one or two mismatches (pink) between g2 and g20. For one mismatch, n = 456: all 19 possible positions × 3 geometries × 4 piRNAs × MILI and MIWI. For two mismatches, n = 1,368: all 171 possible combinations × 1 geometry × 4 piRNAs × MILI and MIWI. Box plots show the IQR and median. Statistical analysesare in Extended Data Fig. 3c. b, Change in pre-steady-state cleavage rate for one or two consecutive mismatches between g2 and g20 for contiguous g2–g21 or g2–g25 pairing for MILI, MIWI, EfPiwi and mouse AGO2. Median and IQR are shown. For one mismatch, n = 24 (3 geometries × 4 piRNAs × MILI and MIWI); n = 6 for EfPiwi (3 geometries × 2 piRNAs); n = 21 for AGO2 (3 geometries for L1MC guide and 3 geometries × 3 contexts for let-7a and miR-21 guides). For two consecutive mismatches, n = 8 (1 geometry × 4 piRNAs × MILI and MIWI); n = 2 for EfPiwi (1 geometry × 2 piRNAs); n = 19 for AGO2 (1 geometry for L1MC RISC and 9 geometries for let-7a and miR-21 RISCs). All data and statistical analyses are in Extended Data Fig. 4. ND, not detected. c, MIWI, MILI and EfPiwi pre-steady-state cleavage rates (k) for targets of L1MC piRNA containing a single unpaired nucleotide. Position and identity of mononucleotide mismatch in targets (indicated in blue) of L1MC piRNA (indicated in red) are on the top of the chart.

Fig. 3. Mouse PIWI proteins cleave partially complementary targets in vivo.

a, Schematic of the strategy used to identify 3′ cleavage products of piRNA-guided PIWI-catalysed slicing and to measure the fraction of targets cleaved by PIWI proteins in FACS-purified mouse primary spermatocytes. b, Fraction of cleaved MILI and MIWI targets in FACS-purified mouse primary spermatocytes for contiguous pairing from nucleotide g2. c, Fraction of cleaved targets in FACS-purified mouse primary spermatocytes for perfect matches (indicated in blue) and for pairing containing a single-nucleotide mismatch (indicated in pink). Horizontal dotted lines indicate the medians for perfect matches. d, MIWI, MILI, and EfPiwi pre-steady-state cleavage rates in vitro for all possible stretches of ≥6-nucleotide contiguous pairing starting from nucleotides g2–g15 of L1MC piRNA. e, Fraction of cleaved targets in FACS-purified mouse primary spermatocytes for 14-nucleotide contiguous pairing starting from nucleotides g2 to g5. Data are binned by piRNA intracellular concentration (<30, 30–50, 50–100, 100–500, >500 pM). For b,c and e, box plots show IQR and median; 95% CI was calculated with 10,000 bootstrapping iterations; n = 16 permutations of 4 control (C57BL/6) and 4 pi2^−/−pi9^−/−pi17^−/− animals.

Fig. 4. Determinants of PIWI slicing in vivo.

a, Decision function coefficients for 400 logistic function fits (regression models) using around 3,500 distinct piRNA–target pairs detected in mouse primary spermatocytes. n = 16 permutations of 4 control (C57BL/6) and 4 pi2^−/−pi9^−/−pi17^−/− animals × 5-repeated × 5-fold cross validation. Box plots show IQR and median. b, Number of piRNAs and siRNAs predicted to cleave mutated versions of the L1Md AI transposon sequence. Data are median and IQR from 100 independent simulations. c, AGO and PIWI proteins use different rules to find and slice targets.

Extended Data Fig. 1. Mouse and freshwater sponge PIWI Argonaute proteins.

a, Overview of RNA Bind-’n-Seq. b, Sequences of small RNA guides used in this study. c, [E]_{active, apparent}, the apparent concentration of active MILI, MIWI, or EfPiwi piRISC, determined for each piRISC purification by fitting the cleavage data (from one experiment) to the burst-and-steady-state equation (see equation and fitting procedure in Analysis of Cleave-’n-Seq Data). d,e, Silver-stained SDS-PAGE gel showing representative purified MILI, MIWI, and EfPiwi piRISCs (d) and Coomassie-stained SDS-PAGE gel showing the purified EmGtsf1 and mouse GTSF1 used in the study (e). For gel source data, see Supplementary Fig. 1.

Extended Data Fig. 2. Binding affinities of Argonaute proteins.

a,b, Mouse and human AGO2 affinities for 6–11-nt complementary stretches contiguously paired from g2 (a) and for 9-nt complementary stretches contiguously paired from all guide nucleotides (b) for miR-34b, miR-449a, let-7, miR-1, miR-7, miR-124, miR-155, and lsy-6. Mouse AGO2 data are from ref. . Human AGO2 data are from ref. . c, Measurements of MIWI piRISC affinity for its targets (0.1 nM) using nitrocellulose filter binding assay. Mean and the standard deviation of the data from three independent trials are shown. d, MIWI, MILI, EfPiwi, and mouse AGO2 binding affinities for ≥11-nt complementary stretches contiguously paired from all guide nucleotides. Mean and standard deviation from three independent trials are shown.

Extended Data Fig. 3. Pairing to piRNA 3′ end is dispensable for PIWI slicing.

a, MILI, MIWI, and EfPiwi pre-steady-state cleavage rates for targets of piRNAs contiguously paired from nucleotide g2. Data are for targets with all possible identities of nucleotide t1.b, Overview of Cleave-’n-Seq. c, Benjamini-Hochberg corrected p-values for post hoc pairwise, two-tailed Mann-Whitney tests for difference in pre-steady-state cleavage rate of targets in Fig. 2a. Kruskal-Wallis test (one-way ANOVA on ranks) p-values are < 10⁻¹⁵ for data with one and two mismatches. d, MILI, MIWI, EfPiwi, and mouse AGO2, binding affinities (K_D) for a g2–g8 match with different t1 nucleotide identities. Mean and standard deviation from three independent trials are shown.

Extended Data Fig. 4. PIWI slicing tolerates mononucleotide and dinucleotide mismatches at any position.

a, Change in EfPiwi, MILI, MIWI, mouse AGO2 pre-steady-state cleavage rate for one or two consecutive mismatches between g2–g20. Box plots show IQR and median: for one mismatch, n = 24 (three geometries × four piRNAs × MILI and MIWI), n = 6 for EfPiwi (three geometries × two piRNAs), n = 21 for AGO2 (three geometries for L1MC guide and three geometries × three contexts for let-7a and miR-21 guides); for two consecutive mismatches, n = 8 (one geometry × four piRNAs × MILI and MIWI), n = 2 for EfPiwi (one geometry × two piRNAs), n = 19 for AGO2 (one geometry for L1MC RISC and nine geometries for let-7a and miR-21 RISCs). b, Benjamini-Hochberg corrected p-values for post hoc pairwise, two-tailed Mann-Whitney tests for difference in pre-steady-state cleavage rate of targets with a mononucleotide mismatch either at g10 or g11 among EfPiwi, MILI, MIWI, mouse AGO2 in panel a. Kruskal-Wallis test (one-way ANOVA on ranks) p-value = 10⁻⁵ for mismatch at g10 and p-value = 4.1 × 10⁻⁶ for mismatch at g11.

Extended Data Fig. 5. PIWI slicing tolerance for different geometries of mismatches.

a, Change in MILI, MIWI, and EfPiwi pre-steady-state cleavage rate for mismatches between g2–g20. Data are binned by mismatch geometry. Data are for all possible mononucleotide mismatch geometries at all 19 positions between g2–g20 for ten piRISCs (Extended Data Fig. 1c). Box plots show IQR and median. Kruskal-Wallis test (one-way ANOVA on ranks) p-value = 1.3 × 10⁻⁶. Benjamini-Hochberg corrected p-values for post hoc pairwise Mann-Whitney tests are shown. b, Change in MILI, MIWI, and EfPiwi pre-steady-state cleavage rate for mismatches at g10 or g11. Data are binned by mismatch geometry. Data are for all possible mononucleotide mismatch geometries at g10 and g11 for ten piRISCs (Extended Data Fig. 1c). Box plots show IQR and the median. Kruskal-Wallis test (one-way ANOVA on ranks) p-value = 0.0004. Benjamini-Hochberg corrected p-values for post hoc pairwise Mann-Whitney tests are shown. c, Relative abundance of 3′ cleavage products generated by L1MC-guided MIWI (2  h at 33 °C). All data and the median from three independent trials are shown. Data are also shown for spike-in RNAs that contained no target sites and were added to the reaction to account for 5′-to-3′ exonucleolytic trimming or non-templated addition of nucleotides to RNA 5′ ends.

Extended Data Fig. 6. GTSF1 accelerates cleavage by PIWI proteins.

a, Pre-steady-state cleavage rate in absence or presence of mouse GTSF1 for mouse AGO2, MILI and MIWI, and EmGtsf1 for EfPiwi. b, Acceleration of pre-steady-state cleavage rates by mouse GTSF1 for MILI and MIWI, and by EmGtsf1 for EfPiwi. Data are for four different piRNA guide sequences bound to MILI or MIWI and for two piRNA guide sequences bound to EfPiwi (Extended Data Fig. 1c). Box plots show IQR and median; 95% confidence interval was calculated with 10,000 bootstrapping iterations.

Extended Data Fig. 7. pi2^−/−, pi7^−/−, pi9^−/−, and pi17^−/− promoter deletions in mice.

Steady-state abundance of piRNA precursor transcripts and mature piRNAs in FACS-purified mouse primary spermatocytes.

Extended Data Fig. 8. PIWI slicing does not require pairing to piRNA 5′ end.

a, Fraction of cleaved targets in FACS-purified mouse primary spermatocytes for pairing containing a single mononucleotide mismatch. Box plots show IQR and median; 95% confidence interval was calculated with 10,000 bootstrapping iterations. b, MILI, MIWI, and EfPiwi pre-steady-state cleavage rates in vitro for all possible stretches of ≥ 6-nt contiguous pairing starting from nucleotides g2–g15 of piRNA #1, Kctd7 piRNA, and piRNA #2. c, Intracellular concentration of pachytene piRNAs in mouse primary spermatocytes. Data are the mean of 12 biologically independent samples.

Extended Data Fig. 9. Target insertions and deletions in the center of piRNA:target duplex are detrimental for PIWI slicing.

a,b, Change in pre-steady-state cleavage rate for mononucleotide target insertions (a) or target deletions (b). Target insertion data are for 16 or 40 targets: four insertion geometries for ten piRISCs (MILI and EfPiwi for pairing up to g26,or MIWI for pairing up to g30). Guide bulge data are for four or ten targets: one deletion geometry for ten piRISCs (MILI and EfPiwi for pairing up to g26 or MIWI for pairing up to g30). Box plots show IQR and median. c, Fraction of cleaved targets in FACS-purified mouse primary spermatocytes for pairing containing a single mononucleotide bulge in target or guide sequence. Box plots show IQR and median; 95% confidence interval was calculated with 10,000 bootstrapping iterations.

Extended Data Fig. 10. Determinants of efficient PIWI slicing in vivo.

a, Area under the precision-recall curves for random control, 400 logistic regression classifier models trained with pi2^−/−; pi9^−/−; pi17^−/− data, and 6,400 tests of the 400 models using pi7^−/− data. Box plots show IQR and median. Kruskal-Wallis test (one-way ANOVA on ranks) p-value = 4.5 × 10⁻¹². FDR (Benjamini-Hochberg) corrected p-values for post hoc pairwise Mann-Whitney tests are shown. b, Number of piRNAs and siRNAs predicted to cleave L1Md_Gf and L1Md_Tf transposon sequences. Data are median and IQR from 100 independent simulations. c, Sequence overlap between transpositionally active families of LINEs and LTR-transposons and exons or introns of mouse mRNAs and lncRNAs.