Uridylation of RNA Hairpins by Tailor Confines the Emergence of MicroRNAs in Drosophila

Frequent Uridylation of Selected miRNAs in Drosophila

To systematically characterize 3′ terminal post-transcriptional modifications of miRNAs we performed high-throughput sequencing of 18- to 30-nt small RNAs of Drosophila S2 cells and adult male flies. Reads were mapped to the genome and separated into genome-matching reads (GM; mapping perfectly to the genome) and prefix-matching reads (PM; containing one or more non-genome-matching nucleotide additions to the 3′ end) (Figure 1A). For all abundantly expressed miRNAs, prefix-matching reads were readily detectable, albeit to varying extent (Figure 1B).

Three criteria classify miRNAs: the relative accumulation of the two strands of a miRNA duplex (miR versus miR^∗), the origin from the 5′ or 3′ arm of a pre-miRNA (5p- versus 3p-miRNA), and the mode of biogenesis by Drosha processing (canonical miRNA) or splicing and debranching (mirtron) (Figure 1C). While the extent to which miRNAs exhibited non-genome-matching nucleotide additions was not consistently different between miR and miR^∗ species, we observed a significantly higher frequency of post-transcriptional modifications at 3p- compared to 5p-miRNAs in both S2 cells and whole male flies. This suggested that pre-miRNAs might be substrates for post-transcriptional modification, which subsequently propagate into mature miRNAs. With a median tailing frequency of close to 10%, mirtrons exhibited the highest modification status, significantly higher compared to canonical miRNAs (p < 1 × 10⁻³ in S2 cells and p < 5 × 10⁻⁴ in whole male flies; Figure 1D). Finally, we determined the nucleotide identity of non-genome-matching additions (Figure 1E): addition of uridine (48% in S2 cells, 52% in flies) and adenine (34% in S2 cells, 42% in flies) dominated over cytosine (10% in S2 cells, 4% in flies) and guanine (8% in S2 cells, 2% in flies). Notably, adenylation was distributed broadly across miRNAs, irrespective of the degree to which these were modified, while uridylation was enriched in highly modified miRNAs (Figure 1E). Frequent uridylation of selected miRNAs suggested a specific targeting strategy for this type of post-transcriptional modification. The predominant class of such miRNAs consists of mirtrons, which are frequently modified (Figure 1D) and predominantly carry uridine modifications (Figure S1D).

The Cytoplasmic TNTase CG1091 Is Required for miRNA Uridylation and Normal Fertility in Flies

The template-independent incorporation of ribonucleoside monophosphate to the 3′ hydroxyl end of RNAs is catalyzed by RNA-specific TNTases. Members of this enzyme family exhibit a characteristic domain architecture consisting of a TNTase domain (NT_PAP_TUTase), often paired with a PAP/25A-associated domain (PAP_assoc) (Norbury, 2013). Based on domain homology search, we identified seven putative TNTases in the Drosophila genome, five of which were expressed in S2 cells (Figure 2A). We depleted candidate enzymes by RNAi in S2 cells followed by detection of the abundantly expressed miR-184 by high-resolution northern hybridization. Although it is not among the most frequently modified miRNAs (∼3%; Figure S1A), miR-184 produced higher-molecular-weight signals in northern hybridization experiments that reflected post-transcriptional modifications in high-throughput sequencing datasets (Figure 2B). Among the tested candidate TNTases, only depletion of CG1091 affected miR-184 tailing (Figure 2C), suggesting that CG1091 is required for the post-transcriptional modification of miR-184.

To characterize the global effect of CG1091 on miRNA modifications we performed high-throughput sequencing of 18- to 30-nt small RNAs from S2 cells depleted of CG1091 by RNAi, as well as adult male flies, carrying two independently generated frame-shift deletions in the third exon of CG1091 (CG1091^c4-1 and CG1091^c4-6; Figure S2A), introduced by CRISPR/Cas9 (Figure 2D). Depletion of CG1091 was confirmed by western blot analysis (Figure 2E). Upon depletion of CG1091, the frequency at which miRNAs contained non-genome-matching nucleotide additions was significantly reduced in S2 cells when compared to control RNAi (p < 10⁻⁴; Wilcoxon matched pairs signed rank test) or untreated cells (p < 3 × 10⁻⁴), and in whole male flies homozygous for CG1091^c-4-1/6 when compared to heterozygous siblings (p < 10⁻⁴) or w¹¹¹⁸ flies (p < 10⁻⁴) (Figures 2F and 2G and Figure S2B). In S2 cells, we detected a statistically significant reduction in the post-transcriptional modification of 48 out of 79 miRNAs (p < 0.05, FDR < 0.1). For 19 of these, tailing was reduced by more than 2-fold (red dots, Figure 2F). In whole male flies, 39 out of 107 miRNAs were consistently reduced in tailing by more than 2-fold in the two independent CG1091^c4 allele-carrying flies (Figures 2F and S2B).

Analysis of the nucleotide identity in miRNA 3′ tails revealed that CG1091 is required for the majority of miRNA uridylation (Figure 2H): upon depletion of CG1091 in S2 cells, uridine-containing extensions decreased by 2.6-fold when compared to the control, while all other nucleotide additions stayed unchanged. Similarly, uridylation decreased by 1.9-fold in CG1091^c4-1/6 heterozygous, and by 4.2-fold in homozygous, flies when compared to w¹¹¹⁸ control flies. Furthermore, depletion of CG1091 affected adenylation, although to a lesser extent (1.2- and 1.4-fold decrease in CG1091^−/+ and CG1091^−/−, respectively), while the addition of cytidine or guanosine was not affected. Note that under CG1091 depletion conditions we still detected residual miRNA uridylation above background, indicating that CG1091 may not be the sole small RNA-uridylating enzyme. CG1091 acts preferentially on mirtrons, which were significantly more depleted in post-transcriptional modifications compared to canonical miRNAs, both in S2 cells (p < 10⁻⁴) and in flies (p < 0.03) (Figure 2I).

While CG1091 is ubiquitously expressed, mRNA levels are highest in the ovary and testis (Graveley et al., 2011), prompting us to investigate any loss-of-function effects on fertility. We observed a significant defect in male (p < 10⁻⁴; Figure 2J) and female (p < 0.01; Figure 2K) fertility in CG1091^c4-1/c4-6 mutant flies when compared to control flies (w¹¹¹⁸). In females, fertility defects were not due to impaired fecundity, since mutant and control flies laid similar numbers of eggs (Figure 2K, top graph). We confirmed the observed phenotypes in trans-heterozygous flies, carrying one CRISPR allele (CG1091^c4-6) over a publicly available piggy-bac transposon insertion in the coding sequence of tailor (CG1091^f05717; Figure 2D) (Thibault et al., 2004), excluding any secondary perturbations linked to the use of CRISPR/Cas9.

Finally, expression of Myc-tagged CG1091 in S2 cells revealed predominant—if not exclusive—localization of CG1091 dispersed throughout the cytoplasm, without detectable signal in the nucleus (Figure 2L and Figures S2D and S2E).

In conclusion, the cytoplasmic, putative TNTase CG1091 is required for normal fertility in flies and is responsible for the majority of uridylation events associated with miRNAs in S2 cells and in flies, with a particular specificity for mirtron uridylation. Because it mediates the selective uridylation of a subset of miRNAs, we refer to CG1091 as Tailor.

Tailor Regulates miRNA Abundance and Function and Prevents the Accumulation of Mirtrons

Post-transcriptional uridylation can trigger RNA destabilization (Scott and Norbury, 2013). We therefore tested if Tailor depletion impacts miRNA abundance. In S2 cells depleted of Tailor, more than 40% of all miRNAs (32 out of 79 miRNAs) exhibited a statistically significant change in abundance (p < 0.05, FDR < 0.1) when compared to control RNAi conditions (Figure 3A). The majority of these (28 out of 32) increased in abundance, suggesting Tailor generally has a negative impact on miRNA accumulation. Mirtrons exhibited a significantly higher accumulation upon Tailor depletion compared to canonical miRNAs (p < 10⁻⁴; Figure 3B), suggesting that Tailor prevents mirton accumulation.

miRNAs that significantly changed in abundance upon depletion of Tailor were more frequently tailed under unperturbed conditions (p < 1 × 10⁻³; Figure 3C and Figure S3A), and tailing of these miRNAs was significantly more dependent on Tailor (p < 0.05; Figure 3D) when compared to miRNAs that remain unchanged in abundance. We concluded that Tailor-mediated uridylation regulates the abundance of miRNAs and prevents the accumulation of mirtrons.

Notably, four miRNAs decreased in abundance upon depletion of Tailor (Figure 3A), including miR-184. We confirmed this effect by northern hybridization, where miR-184 showed a significant, 1.5-fold decrease in abundance upon depletion of Tailor in S2 cells (Figures S3B and S3C). In selected cases Tailor therefore promoted the accumulation of miRNAs.

Finally, high-throughput mRNA sequencing revealed that the observed changes in miRNA levels upon depletion of Tailor significantly impact miRNA-mediated gene regulation (Figures S3D–S3I).

Taken together, Tailor regulates the abundance and function of mature miRNAs and prevents the accumulation of mirtrons.

Tailor-Directed Modification of Pre-miRNA 3′ Ends Regulates miRNA Maturation

To understand how Tailor regulates miRNA abundance we performed gain-of-function experiments by generating clonal S2 cells stably expressing Tailor (CG1091^OE) and monitored miRNA modification and abundance by northern hybridization experiments. Among six tested miRNAs, five significantly changed in abundance upon Tailor expression when compared to control S2 cells (p < 0.05; Figures 4A and 4B): two increased (1.4-fold for both, miR-8-3p and bantam-3p), and three decreased in abundance (2.3-fold for miR-184-3p, 1.2-fold for miR-33-5p, and 3.3-fold for miR-252-5p). But only a subset of the deregulated miRNAs—the ones derived from the 3p arm of the respective pre-miRNA—showed tailing signatures, manifested as higher-molecular-weight bands in northern hybridization experiments (Figure 4A). This indicated that Tailor modifies pre-miRNAs. We tested this hypothesis by analyzing pre-miRNAs in northern hybridization experiments: upon depletion of Dcr-1 by RNAi (enriching pre-miRNAs to detectable amounts) we observed an accumulation of higher-molecular-weight isoforms of both pre-miR-184 and pre-bantam in CG1091^OE cells compared to control S2 cells (Figure 4C). While both pre-miRNAs showed tailing signals, this had differential consequences on the accumulation of the respective mature miRNA species—an increase of bantam-3p and a decrease of miR-184-3p (Figure 4B). Accurate 2-nt 3′ overhangs are hallmarks for pre-miRNA processing by Dicer proteins: changes in pre-miRNA 3′ overhangs impact pre-miRNA processing by Dicer (Tsutsumi et al., 2011). To test if this may explain the observed effects, we cloned and Sanger sequenced pre-miR-184 and pre-bantam to characterize Tailor-directed changes in pre-miRNA 3′ ends (Newman et al., 2011). In control S2 cells, 87% of all pre-miR-184 reads mapped to the predicted 3′ end, producing a 2-nt- 3′ overhang, while the accuracy decreased in CG1091^OE cells to 66%, with 14% containing elongated 3-nt 3′ overhangs (Figure 4D). To test if changes in 3′ end accuracy impact processing, we performed in vitro dicing assays (Tsutsumi et al., 2011): affinity-purified Dcr-1 processed pre-miR-184 with 3-nt 3′ overhang at significantly lower rates when compared to canonical 2-nt 3′ overhangs (p < 10⁻⁴; Figures 4E and S4A–S4C), explaining the observed decrease in miR-184-3p in CG1091^OE cells.

For pre-bantam we observed the opposite: while only 25% of cloned pre-bantam mapped to the predicted 2-nt 3′ overhang in S2 cells (> 40% of all reads supported a blunt 3′ end structure), the accuracy increased in CG1091^OE cells, where 53% of all reads mapped to the predicted 3′ end (Figure 4F). In vitro, Dcr-1-directed processing of pre-bantam was 5-fold more efficient in the presence of a 2-nt 3′ overhang when compared to a blunt end (p < 10⁻³; Figures 4G and S4D–S4F), explaining the increase in mature bantam-3p in CG1091^OE cells (Figure 4B).

We concluded that Tailor-mediated uridylation of pre-miRNAs changes the accuracy of 3′ ends, which impacts the rates at which mature miRNAs are produced by Dcr-1, ultimately regulating the abundance of mature miRNAs.

Tailor-Dependent Uridylation Destabilizes Pre-miRNAs and Prevents Hairpin Dicing

The conclusion that Tailor uridylates pre-miRNAs, as proposed by gain-of-function experiments, warranted further experimental support under conditions where Tailor is expressed at physiological levels. To this end, we cloned 40- to 100-nt-long RNAs from S2 cells using an unbiased cloning approach. As expected, the majority of ∼78 Mio reads in these libraries mapped to abundant, non-coding RNA species that overlapped in size with pre-miRNAs, i.e., tRNA, snoRNA, and rRNA (Figure 5A). While only less than 0.1% of all reads mapped to pre-miRNAs, we recovered sufficient depth (> 70,000 reads) for an unbiased analysis of pre-miRNAs: using a cutoff of > 7 reads, we detected 89 pre-miRNAs, representing the precursors of 73 (50 miR and 23 miR^∗) out of 79 (55 miR and 24 miR^∗) abundantly expressed mature miRNAs. Among the remaining six, three mapped to duplicated miRNA loci and could not be uniquely assigned to one locus in small RNA sequencing datasets (miR-13-b1, miR-276b, and miR-2c); two derived from an unusual, 97-nt-long precursor (miR-998-5p and miR-998-3p), which may have been excluded by size selection; and one was present in small RNA libraries only at low levels (< 250 ppm in small RNA libraries; miR-263a). This analysis confirms that cloning of 40- to 100-nt RNAs at sufficient depth recovers the cellular pre-miRNA pool at near saturation. This conclusion was further supported by the fact that we recovered 38 pre-miRNAs (in average > 50 reads), which were not detected in mature small RNA libraries (Table S1).

For further analysis we focused on the 58 pre-miRNAs that were sequenced at sufficient depth to reproducibly recover species with non-genome-matching 3′ additions (Figures 5B and S5A–S5D). Among these, mirtrons exhibited a significantly higher fraction of non-genome-matching additions compared to canonical pre-miRNAs (p < 10⁻⁴; Figure 5C). Like in mature small RNAs, uridylation (42%) and adenylation (35%) dominated post-transcriptional modifications in pre-miRNAs over cytosine (17%) and guanosine additions (5%) (Figure 5D). Uridylation was likewise enriched in heavily modified pre-miRNAs (Figure 5D) and dominated modifications associated with mirtrons (62%; Figure S5E). Finally, we detected a statistically significant correlation of pre-miRNA tailing and mature miRNA-3p (Pearson’s correlation coefficient r = 0.49; p < 3 × 10⁻³), but not -5p, modifications (r = 0.08, n.s.), indicating that a significant fraction of post-transcriptional modifications detected in mature 3p-miRNAs originated from pre-miRNA tailing (Figure S5F).

To test if Tailor mediates post-transcriptional uridylation of pre-miRNAs we sequenced 40- to 100-nt RNAs from S2 cells depleted of Tailor by RNAi. Tailor depletion caused a significant decrease in tailed pre-miRNAs when compared to control RNAi (p < 0.04; Figures 5E and 5F) or untreated S2 cells (p < 0.03; Figure 5F). Among the non-genome-matching additions to pre-miRNAs, solely uridylation decreased by more than 3-fold upon Tailor depletion when compared to S2 cells treated with control dsRNA (Figure 5G). In comparison to canonical pre-miRNAs, mirtrons were significantly more depleted in post-transcriptional modifications (p < 0.03; Figure 5H).

Finally, we tested the consequences of loss of uridylation on pre-miRNA abundance. We found a significant increase in pre-miRNA levels upon depletion of Tailor for the top 15 (p < 0.02), top 10 (p < 0.02), and top 5 (p < 0.009) Tailor-modified pre-miRNAs (Figure 5J). To determine the impact of uridylation on pre-miRNA processing, we determined in vitro dicing rates of pre-miR-1003, a mirtron that was highly modified in pre-miRNA and mature miRNA libraries (Figures 1B, 2F, and 5B) and was significantly upregulated in its mature form upon depletion of Tailor (Figure 3A): addition of a single uridine to pre-miR-1003 decreased dicing efficiency by more than 2-fold (p < 10⁻⁴), and di- or tri-uridylation reduced processing to almost undetectable levels (p < 10⁻⁸ and p < 10⁻¹¹, respectively; Figures 5J and 5K). Furthermore, pre-miRNA tailing correlated significantly with the observed change in mature miRNA abundance upon depletion of Tailor (Pearson’s r = 0.51, p < 10⁻⁴; Figures S5G and S5H), with a correlation coefficient that was higher than what we observed for the correlation of mature miRNA tailing with change in mature miRNA abundance (Pearson’s r = 0.23, p < 0.05; Figure S3A). Together, these data establish pre-miRNAs—in particular mirtron-hairpins—as major targets for Tailor-dependent uridylation, a modification that destabilizes pre-miRNAs and prevents efficient hairpin dicing.

Tailor Is a Bona Fide, RNA-Specific TUTase with Unique Targeting Properties

For biochemical characterization we immunopurified FLAG-tagged Tailor upon expression in S2 cells and confirmed enzyme purity by western blotting (Figure S6B). When incubated with a 22-nt, 5′ radiolabelled RNA in the presence of ribonucleotide triphosphates (rNTPs), Tailor extended the length of the substrate RNA by up to ∼10 nucleotides within 5 min. This activity was abolished when we replaced one of three highly conserved aspartates in the conserved catalytic pol β superfamily motif with alanine (Figures S6A and S6C). In the presence of individual rNTPs, Tailor catalyzed the efficient incorporation of uridine and—to a lesser extent—cytidine and adenosine, but rarely guanosine (Figures 6A, S6D, and S6E). Tailor-directed terminal nucleotidyl transfer is consistent with the previously proposed two-divalent-cation-dependent catalytic mechanism of TNTases (Martin et al., 2008): the divalent cation-chelating agent EDTA inhibited, while addition of 5 mM excess Mg²⁺ rescued, tailing activity (Figure 6B). Together, these experiments classify Tailor as a bona fide RNA-specific TUTase.

To dissect the enzymatic properties of Tailor in more detail, we developed a high-throughput biochemical assay for the characterization of RNA-specific TNTases. To this end we employed as a substrate a 37-nt RNA containing four random 3′ nucleotides, incubated it with immunopurified Tailor for 2 min or 5 min in the presence of equal concentrations of rNTPs, and subjected the product to high-throughput sequencing (Figure S6F). Analysis of the resulting libraries (∼7 Mio reads per library; Figure S6H) revealed that the tailing product exhibited a length distribution similar to the one observed in denaturing PAGE (Figure 6C). This confirmed a quantitative recapitulation of the tailing reaction by sequencing, and that Tailor catalyzes the incorporation of nucleotides to the 3′ end of an RNA substrate. Notably, Tailor exhibited nearly exclusive selectivity for uridine incorporation: less than 0.02% and 0.25% of all incorporated nucleotides consisted of a nucleotide other than uridine at 2 and 5 min, respectively. The ability of Tailor to efficiently catalyze the incorporation of cytosine or adenosine in the sole presence of CTP or ATP (see Figures S6D and S6E) is therefore apparently without relevance in the presence of UTP.

The randomized 3′ end of the substrate RNA enabled us to simultaneously analyze 256 different substrates, all of which were recovered, and their abundance stayed constant over the course of the assay (Figure S6G). We first determined the secondary structure stability of each substrate, binned them into effective free energy groups (EFE groups) from highly structured (group 1) to single-stranded (group 6), and tested each group for a significant over- or underrepresentation in the mean fraction at which it was tailed (Figure 6D) (note that the substrate was designed to avoid secondary structures not involving the random 3′ nucleotides; see Figures S6I and S6J). This showed that Tailor acted on RNA substrates, irrespective of whether the 3′ end is embedded in secondary structures. In contrast, the number of nucleotides that were added was strongly affected by secondary structures: structured RNAs (EFE groups 1–3) contained significantly shorter tails (p < 10⁻⁴; Mann-Whitney test), whereas single-stranded substrates were tailed significantly longer (EFE group 6; p < 10⁻⁴) (Figure 6E). We concluded that substrate secondary structures impact the processivity of Tailor-directed uridylation.

Finally, we analyzed the impact of primary sequence on tailing efficiency: we found that highly tailed species were enriched in guanosine (Figure 6F). Analysis of each individual position of the randomized four nucleotides revealed that only the 3′ terminal position significantly discriminates tailing efficiency (Figure 6G): while 3′ adenosine or 3′ cytidine substrates were significantly less frequently tailed (p < 10⁻⁴), a 3′ terminal uridine significantly enhanced the tailing reaction (p < 10⁻⁴), and 3′ guanosine-containing substrates were tailed most efficiently (p < 10⁻⁴) at both time points (Figures 6G and S6L). While additional 3′-terminal dinucleotide analysis indicated a trend toward more frequent tailing of substrates ending in AG-3′ (as well as UG-3′ and AU-3′), this further increase in tailing was not statistically significant when compared to the effect of the 3′-terminal nucleotide alone (Figures S6M and S6N).

In summary, high-throughput biochemical characterization of Tailor uncovered substrate structure-dependent processivity and primary sequence specificity, with a preference for RNAs ending in 3′ guanosine (and 3′ uridine). Our data revealed unique targeting properties of a previously uncharacterized RNA-specific TUTase.

Tailor Uridylates Pre-miRNAs Ending in 3′G

To assess primary sequence specificity in Tailor-directed modification of miRNAs, we analyzed the position preceding the non-genome-matching tail in (small) RNA libraries. In both S2 cells and in flies we detected an enrichment of 3′G in Tailor-modified miRNAs (Figure 7A): notably, this effect was most pronounced in pre-miRNAs, the major Tailor substrates.

Selective tailing of substrates ending in 3′G also explained the high degree of Tailor-dependent uridylation of mirtron hairpins, whose 3′ ends are defined by the splice acceptor consensus sequence, ending in 3′ΑG. We confirmed this by in vitro tailing assays using a synthetic precursor of the mirtron miR-1003. While 65% ± 2% of pre-miR-1003 with wild-type sequence (ending in 3′G) was tailed within 15 min, changing the terminal nucleotide to 3′U caused a significant decrease in tailing efficiency (35% ± 2% after 15 min; p < 1 × 10⁻⁴), and mutations to 3′A or C nearly abolished mirtron tailing (9.5% ± 1.8% and 9.3% ± 1.7% tailed, respectively, after 15 min; p < 1 × 10⁻⁵) (Figures 7B and 7C). Single-stranded miR-1003 exhibited a similar 3′ nucleotide preference, albeit with an overall reduced tailing efficiency: while Tailor modified 65% of wild-type pre-miR-1003, only 31% of mature single-stranded miR-1003 was modified within 15 min (Figures S7A and S7B). The intrinsic substrate preference of Tailor for pre-miRNA substrates ending in 3′G therefore explains why mirtron hairpins are preferentially uridylated in flies.

Conserved Pre-miRNAs, but Not Newly Emerging Hairpins, Are Depleted in 3′G

While more than 30 mirtrons have been identified in flies, only few produce abundant, mature miRNAs, perhaps because a mechanism exists that prevents the efficient conversion of such hairpins into small RNAs (Figure 5) (Chung et al., 2011; Westholm and Lai, 2011). We speculated that uridylation of hairpins, such as mirtrons, may represent a regulatory threshold for the accumulation of de novo-generated Dicer substrates. If this were true, one would expect an adaptation to regulatory uridylation in conserved miRNAs. To test this, we investigated the 3′ terminal genome-matching nucleotide of Drosophila pre-miRNAs, expecting that the hallmark of Tailor substrates—a 3′ terminal G—would be underrepresented in conserved pre-miRNAs. We investigated the 3′ end of 143 Drosophila melanogaster miRNAs annotated with high confidence in mirBase, for which sufficient experimental evidence exists to indubitably annotate the 3′ end of the underlying pre-miRNA (Kozomara and Griffiths-Jones, 2014). When compared to the nucleotide distribution in the Drosophila genome, we detected a statistically significant underrepresentation of guanosine at the 3′ end of pre-miRNAs (21% in the genome compared to 15% in pre-miRNA 3′ ends; p < 0.02; binomial test) (Figures 7D and 7E). This depletion was more pronounced in the 108 pre-miRNAs that are conserved among all sequenced Drosophila species, with only 12% of all pre-miRNAs ending in 3′G (p < 0.007). In contrast, among the 35 non-conserved pre-miRNAs, 26% contained a 3′ terminal guanosine, a frequency that was not significantly different from the occurrence of guanosine in the Drosophila genome. Therefore, our data support the hypothesis that evolutionary adaptation to pre-miRNA uridylation shapes the nucleotide composition of pre-miRNA 3′ ends and may serve as a barrier for the efficient de novo creation of hairpins that can be converted into mature miRNAs.

Tailor Targets Non-conserved miRNA Hairpins to Repress Their Maturation

To test if non-conserved miRNA hairpins are more susceptible to Tailor-directed suppression, we determined the extent of post-transcriptional modifications in pre-miRNAs with regard to their evolutionary conservation. We observed a statistically significant, higher fraction of post-transcriptional modifications in poorly conserved pre-miRNAs (i.e., found only in Drosophila melanogaster or Sophophorans) in comparison to miRNA hairpins that were conserved among all 12 sequenced Drosophila species (p < 10⁻⁴; Figure 7F) (Mohammed et al., 2014). Genome-matching reads of non-conserved hairpins tended to end in 3′G more frequently (42%) than conserved hairpins (18%; Figure S7C), consistent with a Tailor-specific targeting strategy. Notably, 14 out of 18 non-conserved hairpins were only detected as pre-miRNAs, but not in mature small RNA sequencing (Figures 7G and S7D), and the few non-conserved miRNAs that we detected in both pre-miRNA and mature miRNA cloning were significantly more abundant upon depletion of Tailor compared to conserved miRNAs (p < 0.03; Figure 7H).

Together our data support the hypothesis that Tailor targets non-conserved miRNA hairpins and prevents their efficient maturation.

General Methods

Total RNA from flies and S2 cells was purified using the mirVana Kit (Ambion) or TRIzol (Invitrogen). High-resolution northern hybridization was as described (Han et al., 2011). Stable and clonal cultured S2 cell lines were generated as described (Ameres et al., 2010). A monoclonal antibody against Tailor was generated against full-length recombinant protein expressed in E. coli. For further information see Supplemental Experimental Procedures.

RNAi in S2 Cells

S2 cells (2 × 10⁶) were soaked with 10 μg dsRNA for 1 hr in ExpressFive SFM medium (Invitrogen) supplemented with 200 mM glutamine (Invitrogen) in the absence of FBS, followed by the addition of medium containing 10% FBS. Soaking was repeated after 4 days, and cells were harvested and processed at day 7.

RNA Library Construction

Small RNA libraries were constructed as described (Ameres et al., 2010) with modifications detailed in Supplemental Experimental Procedures.

In Vitro Tailing Assay

FLAG-tagged Tailor was immunopurified upon expression in S2 cells. In vitro tailing assays employed 10-nM 5′ radiolabelled substrate RNA and FLAG-Tailor under RNAi assays conditions (Ameres et al., 2010), but omitting the ATP-regeneration system. rNTPs were added to a final concentration of 500 μM each. See Supplemental Experimental Procedures for details.

Data Analysis

Gel images were quantified using ImageQuant TL v7.0 (GE Healthcare). Curve fitting was performed according to the integrated rate law for a first-order reaction in Prism v6.0b (GraphPad).

Bioinformatics Analyses and Statistics

For pre-miRNA conservation analysis, high-confidence miRNAs were extracted from mirBase 21 (Kozomara and Griffiths-Jones, 2014) and curated as described in Supplemental Experimental Procedures. Nucleotide composition of the Drosophila genome dm3 was determined excluding chromosome MT, U, and Uextra.

Mapping of sequencing libraries to Drosophila genome dm3 was performed as described (Ameres et al., 2010). Annotations were derived from FlyBase r5.55_FB2014_01. For details see Supplemental Experimental Procedures.

Statistical tests were performed in Prism v6.0b (GraphPad) or Excel v14.4.3 (Microsoft). For multiple hypothesis testing a false-discovery rate (FDR) threshold was set to < 0.1 according to Benjamini and Hochberg (Benjamini and Hochberg, 1995).