Argonaute loading contributes to the precision of the 5′ ends of both microRNAs and their miRNA* strands in flies

Results and Discussion

We used high-throughput pyrosequencing of 18–30 nt RNAs to identify miRNAs expressed in Drosophila melanogaster heads and in cultured Drosophila S2 cells. Among the 120,896 miRNA reads (66,377 from fly heads; 54,519 from S2 cells), we observed two sources of heterogeneity for the ends of fly miRNAs: the addition of nucleotides not present in the gene from which the miRNA is transcribed (non-templated nucleotides) and inaccurate or alternative cleavage by Drosha or Dicer. About 5% of the reads for a typical miRNA contained non-templated nucleotides on at least one end (Figure 2A and S1), most frequently the addition of a single uridine or adenosine to the 3′ end, but longer extensions were also observed, both on the 5′ and the 3′ ends (Table S1). Interestingly, longer extensions were also U- and A-rich at the 3′ end, while at the 5′ end, the 3′-most non-templated nucleotide was frequently a cytidine, and other added nucleotides were typically uridines. This observation could prove useful for the identification of the 5′-elongating enzymatic activity. The non-templated addition of nucleotides, especially uridines, to the 3′ ends of miRNAs has been reported previously in wild-type Caenorhabditis elegans [6] and hen1 mutant Arabidopsis thaliana [10]. Overall, the addition of non-templated nucleotides to the 5′ end of miRNAs was more rare (~1%; Figure 2A and Table S1).

We also observed a second, more frequent type of heterogeneity: variability in the position of the miRNA 5′ and 3′ ends within the sequence of the miRNA precursors (Figure 2B). Non-templated nucleotides fortuitously matching the templated sequence are predicted to occur much less often than the heterogeneity we observe (Table S2). Similar terminal heterogeneity has been noted for the 3′ ends of C. elegans [6] and the 5′ and 3′ ends of mouse [11] miRNAs. The aberrant miRNA termini we observe likely reflect imprecision in precursor cleavage by Drosha and Dicer. They are unlikely to correspond to degradation products, because we recorded nearly as many miRNA reads that were longer than the dominant species as were shorter Figure S1), and because 93% (S2 cells) and 99% (fly heads) of sequences of the fly-specific 30 nt 2S ribosomal RNA (rRNA)—whose termini are expected to be single-stranded—were full-length (Supplemental Discussion). 3′ degradation was slightly more common than 5′ degradation: we detected 3′ degradation for 1,010 reads versus 5′ degradation for 201 reads among the 33,505 total 2S rRNA reads from S2 cells and fly heads combined; 5 reads corresponded to 2S rRNA trimmed from both ends.

The 5′ ends of miRNAs were more precisely defined than their 3′ ends, irrespective of whether the miRNA originated from the 5′ or 3′ arm of the premiRNA (Figure 3A). Thus, the difference in cleavage accuracy between the 5′ and 3′ ends cannot be attributed to an intrinsic difference in fidelity between Drosha and Dcr-1. We expected that the 3′ ends of miRNA* strands would be precisely defined, because they are created by the pair of cuts that generates the 5′ ends of miRNA, and that the 5′ ends of miRNA* strands would be imprecisely determined, because they are created by the pair of cleavages that generates the highly heterogeneous 3′ ends of miRNA. Instead, we found that the 5′ end of a strand (for example, the miRNA) was more accurate than the 3′ end of the adjacent strand (in this example, the miRNA*; Figure 3B); these two extremities are produced by a pair of cuts catalyzed by the same enzyme.

Current dogma holds that the local sequence or structure of miRNA precursors is under strong selective pressure to generate accurate 5′ ends, because a precise miRNA 5′ end directly establishes the seed sequence and hence the targets of the miRNA. Since we observe that, in flies, the 5′ ends of both the miRNA and the miRNA* are more precisely determined than the 3′ ends of either strand, this explanation implies that miRNA* sequences are under selective pressure to establish a unique seed sequence, implying that they, too, have regulatory targets.

It is also possible that both Drosha and Dcr-1—whose active sites are homologous—may also be intrinsically more precise in 5′ cleavage than in 3′ cutting. A third alternative is that 5′ and 3′ ends might be generated with similar, imperfect accuracy, but subsequent constraints in RISC loading or stability select for those small RNAs that begin with a particular nucleotide or sequence. The subsequent destruction of miRNAs without these 5′ features would increase the apparent accuracy of miRNA 5′ ends while retaining miRNA 3′ heterogeneity. To test this idea, we separately sequenced small RNAs containing modified 3′ termini (Table S3). In flies, the 3′ termini of small RNAs that are loaded into Ago2 [12], but not those bound to Argonaute1 [13], are 2′-O-methylated by Drosophila Hen1 as the last step in Ago2-RISC maturation [14]. To sequence small RNAs bearing 2′-O-methylated 3′ ends, we treated the total small RNA with NaIO₄ followed by β-elimination; this method blocks ligation of adapters to small RNAs bearing 2′,3′ hydroxy termini, preventing them from being sequenced.

To determine if the greater accuracy of miRNA and miRNA* 5′ versus 3′ ends reflects the constraints of RISC assembly or stability, rather than more accurate 5′ versus 3′ cleavage by Drosha and Dicer, we compared the terminal heterogeneity of miRNA and miRNA* reads from the 3′ modified population to the heterogeneity of the total miRNA and miRNA* population. As a control, we compared the 3′ heterogeneity between the two populations. For both analyses, we only considered miRNA or miRNA* strands displaying some heterogeneity in the total population. For both fly heads and S2 cells, we observed a dramatic increase in the precision of the 5′—but not the 3′—ends of miRNAs and miRNA* strands upon loading into Ago2 (Figure 4). We also performed the analysis for those small RNAs that both had heterogeneous termini and were specifically enriched in the b-eliminated sequences relative to the non-b-eliminated set. For the 13 small RNAs (4 miRNAs and 9 miRNA*s) meeting these criteria, the 5′ ends in the sub-population of miRNA and miRNA* sequences loaded into Ago2—i.e., those that were 2′-O-methylated—were again more precisely defined than the 5′ ends of the same small RNA sequences in the total small RNA population (Figure S3). We conclude that loading or stabilization of miRNAs in Ago2, and, perhaps Argonaute proteins in general, imposes a purifying selection on their 5′ ends.

The mechanism responsible for the homogenization of 5′ ends remains to be determined. We can imagine that the efficiency of Argonaute loading is affected by the nature of the 5′ end of a small RNA, much as the stability of its pairing to the other strand influences this process [21]. The 5′ sequence itself may also play a role in RISC assembly, with some miRNA variants loaded more efficiently than others, according to the identity of their 5′ nucleotide(s). Alternatively, some Argonaute complexes might be selectively stabilized after their assembly, for example, by the presence of a target RNA whose binding stabilizes those RISCs containing miRNA isoforms with a complementary seed sequence.

Experimental Procedures

Supplementary Material