Abstract
MicroRNAs (miRNAs) exert their gene regulatory effects on numerous biological processes based on their selection of target transcripts. Current experimental methods available to identify miRNA targets are laborious and require millions of cells. Here we have overcome these limitations by fusing the miRNA effector protein Argonaute2 to the RNA editing domain of ADAR2, allowing the detection of miRNA targets transcriptome-wide in single cells. miRNAs guide the fusion protein to their natural target transcripts, causing them to undergo A>I editing, which can be detected by sensitive single-cell RNA sequencing. We show that agoTRIBE identifies functional miRNA targets, which are supported by evolutionary sequence conservation. In one application of the method we study microRNA interactions in single cells and identify substantial differential targeting across the cell cycle. AgoTRIBE also provides transcriptome-wide measurements of RNA abundance and allows the deconvolution of miRNA targeting in complex tissues at the single-cell level.
Main
miRNAs are small noncoding RNAs that posttranscriptionally regulate the expression of protein-coding genes1. Mechanistically they guide Argonaute effector proteins to messenger RNA targets, allowing Argonaute and cofactors to inhibit translation and/or promote degradation of target mRNAs2,3,4,5. miRNAs are found in virtually all multicellular animals and plants and play important roles in numerous biological processes, including development, formation of cell identity and human diseases such as cancer1,6,7,8. The human genome harbors hundreds of distinct miRNA genes9, each of which can putatively regulate hundreds of target genes. The function of each individual miRNA is defined by its specific target repertoire and thus, to understand the function of a given miRNA, it is necessary to map its targets. The current state-of-the-art method to do so is crosslinking-immunoprecipitation-sequencing (CLIP-seq), which applies ultraviolet light to crosslink the Argonaute protein to its mRNA targets in cells, then isolates the protein using antibodies and uses next-generation sequencing to profile the bound RNA targets10,11. This method has brought many new insights to the miRNA field yet it has some inherent limitations. First, because isolation with antibodies is inefficient, it requires as input in the order of millions of cells, making it unsuited for samples with limited material—not to mention in single cells. By averaging over millions of cells, CLIP-seq masks potential cell-to-cell variation. Second, the method is laborious and requires many specialized protocol steps, including ultraviolet crosslinking and immunoprecipitation.
We here present our method agoTRIBE, which circumvents these limitations. We show that our method yields results that are consistent with the more laborious CLIP-seq method. The reported miRNA targets are supported by evolutionary sequence conservation and are subject to functional miRNA repression. In addition, we show that agoTRIBE can be applied to the detection of miRNA–target interactions in human single cells, and to deconvolution of miRNA targeting in distinct phases of the cell cycle without the need for physical cell sorting.
Results
To develop our method we leveraged on the TRIBE approach12, in which an RNA-binding protein of interest (in our case, Argonaute2) is fused to the RNA editing domain of ADAR2. The RNA-binding protein part leads the fusion protein to its natural targets while the editing domain deaminates adenosines to inosines (A>I) in the RNA target, in effect leaving nucleotide conversions that can be detected by sequencing as A>G substitutions (Fig. 1). These substitutions can in principle be detected by single-cell RNA sequencing (RNA-seq), and the method avoids lossy isolation because it does not use immunoprecipitation. To tailor agoTRIBE for Argonaute proteins we made three modifications to the original TRIBE approach: (1) we used a hyperactive version of the ADAR2 deaminase domain, in which a E488Q substitution results in increased editing13,14; (2) we connected the Argonaute2 and ADAR2 domains with a 55-amino-acids-long flexible linker; and (3) we fused the ADAR2 domain to the N terminus of Argonaute2, because the protein structure of Argonaute2 indicates that fusion to the C terminus would be detrimental to the loading of guide miRNAs15,16,17,18,19 (Fig. 1b). We confirmed that tagging Argonaute2 with the ADAR2 editing domain does not change its cytoplasmic localization (Fig. 1c), nor its colocalization with TNRC6B, a P-body marker (Supplementary Fig. 1a). In particular, we found that agoTRIBE partly locates to cytoplasmic foci that are similar to P-bodies, which are known to be interconnected with miRNA function20,21,22 (Fig. 1c).
When we transiently expressed agoTRIBE in ~50,000 human HEK-293T cells (Methods), we observed that A>G nucleotide substitutions—as expected by ADAR2-mediated editing—increased substantially compared with control cells where editing is low, consistent with ADAR2 being undetectable in this cell type (Fig. 2a, Supplementary Tables 1–4 and Supplementary Fig. 1b–d). In contrast, cells transfected with only the ADAR2 deaminase domain without Argonaute2—henceforward referred as ‘ADAR-only’—increased only moderately the number of A>G substitutions, suggesting the importance of miRNA guidance for newly detected editing (Fig. 2a). Of note, other types of nucleotide substitution remained largely unchanged among the analyzed conditions, indicating specific ADAR2-mediated editing. In addition, we observed that editing in mRNA exonic regions specifically increased while editing in intronic regions and noncoding transcripts such as long noncoding RNAs and pseudogenes remained constant (Fig. 2b,c). This is consistent with miRNAs targeting mature mRNAs in the cytoplasm, while there is little evidence of miRNAs targeting nontranslating sequences such as introns or lncRNA transcripts23, which are most commonly located in the nucleus. We note that the agoTRIBE construct did not appear to substantially change the miRNA profile of transfected cells (Supplementary Fig. 2). In summary, we observed highly increased editing in mRNA transcripts that likely correspond to natural miRNA targets…
