MiRNAs have emerged as one of the key regulators for gene expression. Before isomiRs were discovered, the miRNA variants were usually missed or ignored by traditional miRNA cloning technique. With the advances in deep sequencing, increasing numbers of miRNAs and its cognate miRNAs, miRNA-3p, were found to differ from the currently annotated sequence in miRBase, and the population of miRNA isoforms varied among different tissues or cell types 
. However, the possibility of concordant or discordant regulation of target genes by different isoforms of miRNAs has not been validated at the cellular level until this report. In this study, we used miR-31 as a model to demonstrate that the most abundant isoform of miR-31 and its cognate miRNA, miR-31-3p, varied in different cells by comparing our deep sequencing data in MCF-7, HCT116, and LNCaP with the previous report (hES and hEB) ( and Figure S5
). We further investigated the functions of isomiRs at the cellular level and provided direct evidences that isomiRs are not equal in their target regulation. Previously, it was reported that hundreds of mRNA enriched in the miRNA pull-down were common to their isomiR pull-downs by microarray analysis 
. However, a close scrutiny of their data revealed that some mRNA targets were unique to the specific isomiRs. Such systems analysis, although powerful, did not offer direct proof for the regulation of a particular target by specific isomiRs. Herein, our studies have provided solid evidence for the complexity of target regulation by isomiRs at the cellular level.
Several inherent challenges in the investigation of isomiRs were encountered in our study. First, traditional cloning and sequencing is not ideal for quantifying isomiRs because cloning frequencies may not truly reflect the isomiR populations. Another technical limitation of traditional cloning is to accurately delineate 5′- or 3′-end sequence information of a specific miRNA (see Figure S6
). The use of northern blot analysis is not practical for isomiR study either, since there are no available commercial probes including LNA detection probe that can guarantee specific distinction of our three isomiR-31s. Even if miR-31-M and miR-31-H/−P were distinguishable by northern blotting, it is not possible to separate miR-31-H and miR-31-P from each other because of their identical length. Although the TaqMan qPCR probes were widely used in miRNA studies, we found that the same probe for miR-31-H could also recognize the other two isoforms (Figure S4D
). Thus, the specificity of the TaqMan probe is not sensitive enough for our experiments. Hence, deep sequencing is the only reliable approach to identify the endogenous isomiRs populations in different cells or tissue. The second challenge is the limited choice of strategies for overexpressing and silencing specific isomiRs. Since isomiRs were processed from the same pri-miRNA/pre-miRNA, it will not be straightforward to identify specific isomiR-31 generated by transfecting cells with a plasmid bearing pri-miR-31/pre-miR31 sequence, making it difficult to attribute the observed phenotype to any specific isoform after transfection. Instead, we used synthetic double stranded miRNA oligos pledged by Ambion and Dharmacon for transfection into cancer cells to compare the functions of isomiRs. To further confirming our finding by silencing a specific isoform of miR-31 is not feasible either, because of a lack of molecules that are guaranteed to inhibit specific endogenous miRNA isoform. Thus, to address the functions of isomiRs in depth, it may be necessary to simultaneously decipher the expression profile of target genes and the populations of isomiRs in different types of cells, which awaits future studies.
Since gene regulation mediated by miRNA requires the ternary interactions among miRNA/AGO/target mRNA, it is possible that differential interactions of isomiRs within the ternary complex may lead to disparate regulation of target genes. In this study, the observed discrepancy between the miR-31 isoforms bound within AGO-IP and their repression of Dicer and other known target genes suggested that the affinity of a given miRNA to AGO or their seed sequences might not be the only critical elements for the target gene repression. In fact, several factors have been shown to dictate the recognition of target site by miRNA, such as (1) the sequence composition of the 3′-UTR 
, (2) the immediate environment of the putative target site 
, (3) the structural accessibility of the target site 
, and so forth. Besides, endogenous natural antisense transcript (NAT), which was transcribed from the opposite strand of protein-coding gene or non-protein coding gene 
, and the RNA binding proteins 
could directly bind to mRNA, thereby masking the miRNA binding site of target gene and preventing the inhibitory effects of the miRNA on target gene translation. Although the bindings of miR-31-P and –M to AGO complexes were comparable, the above-mentioned factors might come into play in the differential regulation of Dicer and other known target genes expression. The exact mechanisms underlying the target specificity of isomiRs await further investigation in the future. Taken together, the variations in the relative abundance of isomiRs among different cell types coupled with our finding that isomiRs could differentially regulate the expression of target genes, suggest that isomiRs may play a more general and weighty role in nature by fine-tuning target gene expression.