Twenty-four-nucleotide small interfering (si)RNAs are central players in RNA-directed DNA methylation (RdDM), a process that establishes and maintains DNA methylation at transposable elements to ensure genome stability in plants. The plant-specific RNA polymerase IV (Pol IV) is required for siRNA biogenesis and is believed to transcribe RdDM loci to produce primary transcripts that are converted to double-stranded RNAs (dsRNAs) by RDR2 to serve as siRNA precursors. Yet, no such siRNA precursor transcripts have ever been reported. Here, through genome-wide profiling of RNAs in genotypes that compromise the processing of siRNA precursors, we were able to identify Pol IV/RDR2-dependent transcripts from tens of thousands of loci. We show that Pol IV/RDR2-dependent transcripts correspond to both DNA strands, whereas the RNA polymerase II (Pol II)-dependent transcripts produced upon derepression of the loci are derived primarily from one strand. We also show that Pol IV/RDR2-dependent transcripts have a 5' monophosphate, lack a poly(A) tail at the 3' end, and contain no introns; these features distinguish them from Pol II-dependent transcripts. Like Pol II-transcribed genic regions, Pol IV-transcribed regions are flanked by A/T-rich sequences depleted in nucleosomes, which highlights similarities in Pol II- and Pol IV-mediated transcription. Computational analysis of siRNA abundance from various mutants reveals differences in the regulation of siRNA biogenesis at two types of loci that undergo CHH methylation via two different DNA methyltransferases. These findings begin to reveal features of Pol IV/RDR2-mediated transcription at the heart of genome stability in plants.

3' uridylation is increasingly recognized as a conserved RNA modification process associated with RNA turnover in eukaryotes. 2'-O-methylation on the 3' terminal ribose protects micro(mi)RNAs from 3' truncation and 3' uridylation in Arabidopsis. Previously, we identified HESO1 as the nucleotidyl transferase that uridylates most unmethylated miRNAs in vivo, but substantial 3' tailing of miRNAs still remains in heso1 loss-of-function mutants. In this study, we found that among nine other potential nucleotidyl transferases, UTP:RNA uridylyltransferase 1 (URT1) is the single most predominant nucleotidyl transferase that tails miRNAs. URT1 and HESO1 prefer substrates with different 3' end nucleotides in vitro and act cooperatively to tail different forms of the same miRNAs in vivo. Moreover, both HESO1 and URT1 exhibit nucleotidyl transferase activity on AGO1-bound miRNAs. Although these enzymes are able to add long tails to AGO1-bound miRNAs, the tailed miRNAs remain associated with AGO1. Moreover, tailing of AGO1-bound miRNA165/6 drastically reduces the slicing activity of AGO1-miR165/6, suggesting that tailing reduces miRNA activity. However, monouridylation of miR171a by URT1 endows the miRNA the ability to trigger the biogenesis of secondary siRNAs. Therefore, 3' tailing could affect the activities of miRNAs in addition to leading to miRNA degradation.