MicroRNAs and their isomiRs function cooperatively to target common biological pathways

Abstract

Background: Variants of microRNAs (miRNAs), called isomiRs, are commonly reported in deep-sequencing studies; however, the functional significance of these variants remains controversial. Observational studies show that isomiR patterns are non-random, hinting that these molecules could be regulated and therefore functional, although no conclusive biological role has been demonstrated for these molecules.

Results: To assess the biological relevance of isomiRs, we have performed ultra-deep miRNA-seq on ten adult human tissues, and created an analysis pipeline called miRNA-MATE to align, annotate, and analyze miRNAs and their isomiRs. We find that isomiRs share sequence and expression characteristics with canonical miRNAs, and are generally strongly correlated with canonical miRNA expression. A large proportion of isomiRs potentially derive from AGO2 cleavage independent of Dicer. We isolated polyribosome-associated mRNA, captured the mRNA-bound miRNAs, and found that isomiRs and canonical miRNAs are equally associated with translational machinery. Finally, we transfected cells with biotinylated RNA duplexes encoding isomiRs or their canonical counterparts and directly assayed their mRNA targets. These studies allow us to experimentally determine genome-wide mRNA targets, and these experiments showed substantial overlap in functional mRNA networks suppressed by both canonical miRNAs and their isomiRs.

Conclusions: Together, these results find isomiRs to be biologically relevant and functionally cooperative partners of canonical miRNAs that act coordinately to target pathways of functionally related genes. This work exposes the complexity of the miRNA-transcriptome, and helps explain a major miRNA paradox: how specific regulation of biological processes can occur when the specificity of miRNA targeting is mediated by only 6 to 11 nucleotides.

Figure 1

Correlation between miRNA-seq, qRT-PCR, and microarrays for placental miRNAs. Different alignment strategies will produce different correlations due to the characteristics of the validation technology. (a) The Agilent miRNA microarray typically will capture signal from isomiRs due to cross-hybridization, and therefore has a higher correlation to the recursive matching strategy, which identifies both miRNAs and isomiRs. (b) TaqMan LDA qRT-PCR is more sensitive to the exact sequence of the miRNA, and therefore has higher correlations to the adaptor trimming strategy, which was filtered for exact matches to canonical miRNA sequences. In all plots, points represent mean detection values, and bi-directional error bars represent standard error. ρ values indicated are Spearman correlations. These comparisons demonstrate that the miRNA-seq technology and our analysis approach are both sensitive and specific.

Figure 2

Arm switching frequency in the miRNA-seq data. (a) Arm-specific expression of at least 21 miRNAs varied between tissues. For all 10 tissues (points), log ratios of 5p to 3p arm expression for 21 miRNAs (blue lines) are plotted according to the arm on which dominant expression occurred, clearly illustrating differences between the tissues for each miRNA. (b) Sequence logo plots for hsa-miR-126 showing the switch between expression dominance from the 5p arm (placenta) to the 3p arm (heart, liver, and kidney). The pre-miRNA hairpin is displayed along the bottom.

Figure 3

miRNA and isomiR characteristics. (a) Example of a canonical miRNA and each mutually exclusive isomiR category. IsomiRs that would fit more than one of these categories are classified as 'mixed' isomiRs. (b) Box-whisker plots showing expression of isomiR types within each tissue. (c) Ratio of each isomiR to its canonical miRNA grouped according to isomiR type within each tissue.

Figure 4

Identification of ac-pre-miRNA candidates. (a) Sequence logo profile of hsa-miR-451 showing 5' arm dominance, almost no variation in the 3' end of the mature miRNA, and a trail of 3' isomiR extensions. (b) Schematic diagram illustrating the generation of end-site (5p-ac-pre-miRNA) and start-site (3p-ac-pre-miRNA) isomiRs via AGO2-mediated cleavage of a pre-miRNA. Dashed lines indicate variable regions. (c) Sequence logo profile for hsa-miR-31, which was identified by our analysis as a candidate for biogenesis through the ac-pre-miRNA pathway. Three representative tissues are shown. (d) Wild-type (WT) zebrafish whole embryo versus mutant AGO2 inactive (MZAgo2) whole embryo miRNA and isomiR expression. Each point represents a separate isomiR or miRNA. The thick blue line is the perfect correlation line, the thin blue lines are ± 2-fold visual guides. Red dots indicate isomiRs/miRNAs derived from candidate ac-pre-miRNA hairpins. A bias towards down-regulation in AGO2 mutants is evident in this plot. (e) Box and whiskers plots showing the expression of all isomiRs/miRNAs deriving from ac-pre-miRNA candidates in wild-type (WT) and mutant (MZago2) miRNA-seq samples; 75% of ac-pre-miRNA candidate molecules are not expressed in the mutant sample.

Figure 5

IsomiRs are associated with translational machinery. (a) Schematic diagram showing the laboratory workflow to isolate total miRNA, miRNA associated with mRNA, and miRNAs associated with polysomal mRNA. (b, c) Correlation scatter plots describing miRNA expression in the three preparations described in (a). The noise threshold was set at 25 tpm for these libraries. (b) Correlation scatter plot between miRNA associated with mRNA (x-axis), and miRNA associated with polysomal mRNA (y-axis). The dashed diagonal line represents perfect correlation, while the horizontal and vertical lines represent the likely noise thresholds for these data. (c) Correlation scatter plot between total cellular miRNAs, and polysomal associated miRNA. The dashed diagonal line represents perfect correlation, while the horizontal and vertical lines represent the likely noise thresholds for these data. (d) A sequence logo profile for hsa-miR-31 showing similar profiles for all three HeLa miRNA-seq samples. This example shows an ac-pre-miRNA candidate hairpin with no change in the proportion of intermediates in the translational machinery. (e) Proportional bar graph illustrating the similar distribution of isomiR categories in each of the three HeLa miRNA-seq samples. CHX, cycloheximide, nt, nucleotide.

Figure 6

IsomiRs target similar molecules to their canonical miRNAs. (a) Schematic diagram depicting the laboratory workflow for the pull-down of mRNA targets of a synthetic biotinylated miRNA. (b) The miRNA expression profiles of the two miRNAs and corresponding shifted isomiRs selected for pull-down experiments. (c) A four-way Venn diagram depicting the overlap between the target genes detected by microarray analysis of pulled-down fractions. (d) The distribution of the number of miRNAs/isomiRs that target genes from significantly enriched pathways. There are more genes targeted by three or more isomiRs than would be expected by chance, and less genes targeted by two or fewer isomiRs than would be expected by chance. (e) A four-way Venn diagram displaying the specific targeting of the genes used in (d).

Figure 7

IsomiRs and miRNAs target similar biological pathways. The 'molecular mechanisms of cancer' pathway (Ingenuity Pathway Analysis) overlaid with rectangles corresponding to the miRNA or isomiR targeting that gene. hsa-miR-10a-5p|{hsa-miR-10a}|21_43| is colored red, hsa-miR-10b-5p|{hsa-miR-10b}|26_48|) is colored yellow, hsa-miR-10a-5p|{isomiR}|22_44| is colored blue, and hsa-miR-10b-5p|{isomiR}|27_49| is colored green.

Figure 8

A diagram illustrating how isomiRs may act cooperatively with the canonical miRNAs to increase the signal to noise ratio of mRNA targeting. (a, b) Where only a single miRNA molecule is used for targeting (a), there is no differentiation between potential 'off-target' effects, and targeting the 'desired' core molecular network when the dosage of that miRNA is increased. However, the dosage is increased using isomiRs (b), with slightly different targeting of both core network mRNAs and 'off-target' miRNAs, amplification of specific targeting of the 'core' network can be achieved relative to the 'off-target' effects. The more isomiRs present, the more differentiation could be achieved.