The regulatory mechanisms of miRNA-mediated regulation present a complex picture. Initially discovered to inhibit protein translation, miRNAs were later discovered to downregulate mRNA transcript levels of their target mRNA. Further studies showed that in the right context, they could function to activate protein translation. Because miRNAs regulate their targets via an imperfect binding of the 22 nucleotide molecule to their target mRNAs, each miRNA has the potential to directly regulate hundreds of mRNA transcripts. Conversely, the length of most mRNA transcripts provide potentially many binding sites for the small non-coding regulatory molecule and miRNAs have in fact been shown to act cooperatively to regulate a given transcript (Saetrom et al, 2007
; Hobert, 2008
). miRNAs have also been theorized to act as buffers in regulating mRNA transcripts (Hornstein and Shomron, 2006
; Herranz and Cohen, 2010
In addition to the complexity of miRNA–mRNA relationships, there are numerous mechanisms regulating both overall content of miRNAs (e.g. factors involved in miRNA biogenesis) and specific miRNAs themselves. For example, uridylation of mature miR-26a has been shown to reduce miR-26a activity without significantly affecting its expression levels, while uridylation of the precursor form of let-7a targets the miRNA precursor for degradation (Kai and Pasquinelli, 2010
). Another study demonstrated the buffering effect of mRNA from psudeogenes against miRNA activity (Poliseno et al, 2010
). Here, the pseudogene PTENP1 was shown to act as a decoy for miRNAs targeting PTEN with decreasing levels of PTENP1 leading to increase activity of PTEN-targeting miRNAs on the PTEN transcript (Poliseno et al, 2010
). These studies suggest not only complex regulation of biological pathways via miRNAs, but also intricate regulation of miRNAs transcript abundances themselves and similarly complex regulation of miRNA activity.
Our findings in this paper provide further support and characterization of the complexity of miRNA-mediated regulation in gene regulatory networks. Using an integrative genomics approach, we have explored the genetic basis of miRNA expression variation for a substantial fraction of known miRNAs (~30% of those registered in mirBase (version 15)). Despite surveying approximately a third of known mouse miRNAs today, many of the surveyed miRNAs were expressed in liver. We showed that variation in transcript abundances of many miRNA are linked to DNA sequence variation and that, in a number of ways, the genetics of miRNA gene expression is similar to mRNA expression traits. Most notably, the strongest LOD scores for miRNA expression traits are associated with cis-acting miRNA eQTLs, which is a frequent observation with mRNA expression traits. In addition, we were similarly able to detect the presence of miRNA eQTL hotspots, that is, regions that account for the trait variation of a disproportionally larger number of expression traits than would be expected by chance.
Because only a handful of miRNA eQTLs were detected using a genome-wide scan, we postulate that the effects of DNA variation on miRNAs are more subtle than the effects of sequence variation on mRNA transcripts. In this case, each miRNA may be regulated by multiple eQTLs, each with an effect size that is typically smaller than that detected for mRNA eQTLs. Alternatively, it is also possible that the most miRNAs are simply not affected by polymorphisms presence in the mouse genome or that the proportions observed in this study are simply a chance occurrence unintentionally resulting from surveying only a third of miRNAs that have been discovered. The former explanation would be concordant with miRNAs being regulated by an intricate network of multiple genes.
To address this ambiguity, we restricted the potential loci to those that only encompassed eQTLs for the set of potential target mRNAs of each miRNAs and noted an increase in statistical power to detect miRNA eQTLs. Overall, we were able to detect more than a seven-fold increase in miRNA eQTLs strongly suggesting that the variation in miRNA expression trait that is accounted for by DNA sequence variation is indeed smaller than that of mRNA expression trait. From the standpoint of gene regulatory networks, this observation supports the notion of complex interplay between miRNAs and the genes located in those regions.
Of the set of miRNA eQTLs detected, 85% were trans
eQTLs, suggesting that many miRNAs are subject to controlled extending beyond local sequence variants. In the simplest model, this could be viewed as a perturbation to an element or gene contained within the trans
locus that indirectly then influences miRNA expression trait levels. Recent studies on trans-
splicing events and the prevalence of chimeric RNA suggests alternative implications for trans
eQTLs, where such events may directly lead to increase or decrease in miRNA transcript levels (Gingeras, 2009
In examining the correlation structure between miRNA and mRNA, we noticed that counter to the widespread acceptance that miRNAs downregulate gene expression, a surprisingly large proportion of miRNAs signature sets (i.e. sets of mRNA transcripts that are significantly correlated with a given miRNA) that were negatively correlated with their respective miRNA showed poor enrichment for the corresponding miRNA seed in the 3′ UTR of the mRNA transcripts within the set. Instead, many miRNAs signature sets that displayed a positive correlation with the individual miRNAs were observed to be enriched for the seed region of each respective miRNA in the 3′ UTR of the mRNA. While unexpected, strong positive correlations between miRNAs and mRNAs have been previously reported in a number of studies (Liu et al, 2007
; Tsang et al, 2007
; Nunez-Iglesias et al, 2010
). Additionally, using the expression of host genes with embedded miRNAs as a proxy for expression of the embedded miRNA, Tsang et al (2007)
found significant positive and negative correlation between the embedded miRNAs and targets of the embedded miRNAs as predicted by TargetScanS algorithm, an miRNA prediction algorithm that utilizes the conserved Watson–Crick pairing (i.e. miRNA seed) between the miRNA and its target mRNA (Tsang et al, 2007
One possible explanation for the occurrence of positive correlation between a given miRNA and their respective target that has been previously been postulated is the presence of feedback motifs. Feedback motifs are known to be common within gene regulatory networks and studies specific to miRNA–mRNA networks have shown an enrichment of feedback loops (Martinez et al, 2008
). Others have elucidated and classified specific feedforward and feedback circuit motifs that could explain the both positive and negative miRNA target mRNA correlations (Tsang et al, 2007
). While examples of miRNA activating transcription are not yet widespread, this represents another plausible explanation for our observation.
Further analysis of the causal relationship between miRNAs and mRNAs suggests that for many miRNAs, the set of protein-coding genes predicted to regulate levels of a given miRNA exceeds that of its downstream target mRNAs. Simulation studies that take into account the difference in noise levels between the two data sets indicate that the number of predicted miRNA targets is likely to be an overestimation while the number of predicted miRNA regulators is likely to be an underestimation. Overall, these results add to the increasing complexity of miRNA-mediated regulation.
Using GO Biological Process and KEGG pathways, we show that almost a quarter of the surveyed miRNAs are correlated with mRNA transcripts involved in DNA replication and cell-cycle regulation within the context of a liver gene regulatory network. These miRNAs are often correlated with the sets of mRNA transcripts that are significantly associated with one another, suggesting that these miRNAs cooperatively or redundantly act to regulate a core set of mRNA transcript within a given pathway. In a few cases, two or more miRNAs were significantly correlated with two distinct sets of mRNA transcripts belonging to the same pathway.
To our knowledge, our results represent one of the first examinations of the genetic basis of variation in miRNA expression and exemplify how integrative genomics approaches may be used to elucidate multiple insights surrounding the regulatory circuitry of novel classes of RNA molecules. With advances in next-generation sequencing technologies, this approach can be extended to characterize the full set of transcriptional units (i.e. non-coding RNA, chimeric RNA, mRNA etc.) to decipher the myriad of molecular regulatory relationships that governs phenotypic behavior. From our study, given the abundance of miRNA–mRNA interaction, it seems that miRNAs may be sensing network states and responding to entire network changes in mRNA levels. We hypothesize that miRNAs then respond in a programmed manner to drive pathway changes via modulation of specific sets of mRNA.