Complexity of microRNA function and the role of isomiRs in lipid homeostasis.

MicroRNAs (miRNAs) are key posttranscriptional regulators of biological pathways that govern lipid metabolic phenotypes. Recent advances in high-throughput small RNA sequencing technology have revealed the complex and dynamic repertoire of miRNAs. Specifically, it has been demonstrated that a single genomic locus can give rise to multiple, functionally distinct miRNA isoforms (isomiR). There are several mechanisms by which isomiRs can be generated, including processing heterogeneity and posttranscriptional modifications, such as RNA editing, exonuclease-mediated nucleotide trimming, and/or nontemplated nucleotide addition (NTA). NTAs are dominant at the 3'-end of a miRNA, are most commonly uridylation or adenlyation events, and are catalyzed by one or more of several nucleotidyl transferase enzymes. 3' NTAs can affect miRNA stability and/or activity and are physiologically regulated, whereas modifications to the 5'-ends of miRNAs likely alter miRNA targeting activity. Recent evidence also suggests that the biogenesis of specific miRNAs, or small RNAs that act as miRNAs, can occur through unconventional mechanisms that circumvent key canonical miRNA processing steps. The unveiling of miRNA diversity has significantly added to our view of the complexity of miRNA function. In this review we present the current understanding of the biological relevance of isomiRs and their potential role in regulating lipid metabolism.

independently along multiple lineages. A recent large-scale meta-analysis of hundreds of smRNA data sets has systematically identifi ed ‫ف‬ 240 human miRtrons, many of them evolutionarily recent ( 76 ). Finally, one highly conserved pre-miRNA (pre-miR-451) has been demonstrated to bypass DICER processing altogether and is instead cleaved directly by Argonaute 2 (Ago2) within the RISC (79)(80)(81).
MIRNAS CONTROL HEPATIC LIPID METABOLISM miRNAs bind to target RNAs through complementary base pairing, most critically in the seed region of the miRNA (nucleotides 2 through 7 at the 5 ′ -end). One of the fi rst miRNAs shown to have a role in lipid biology is miR-33, which is cotranscribed with the SREBF family of genes and is a critical regulator of lipoprotein metabolism and fatty acid oxidation ( 44,46,49,82 ). The functional relevance of miR-33 to lipid homeostasis has been established in both mice and in nonhuman primates with systemic administration of chemically modifi ed locked nucleic acid (LNA), which mediates safe and potent inhibition of endogenous miRNAs in vivo ( 46,55,83 ). Most recently, miR-33 has also been suggested as a key link between lipid metabolism and hepatocyte cell proliferation, particularly in the context of response to liver injury ( 47,84 ).
Using in silico strategies for miRNA target prediction and enrichment analysis, we recently identifi ed several miRNAs, including miR-27, miR-223, and miR-125, as candidate regulatory hubs in the pathways that govern lipid homeostasis ( 85 ). We demonstrated in vitro in human hepatocytes that miR-27b represses the expression of glycerol-3-phosphate acyltransferase 1 (GPAM), angiopoietin-like 3 (ANGPTL3), peroxisome proliferator-activated receptor ␥ (PPAR ␥ ), and heparan sulfate N-deacetylase/ N-sulfotransferase 1 (NDST1) ( 15,(85)(86)(87)(88)(89)(90)(91). Inhibition of miR-27b in hepatocytes resulted in a signifi cant increase in GPAM mRNA and protein levels. GPAM-mediated acyl esterifi cation of glycerol-3-phosphate (sn-1 position) is the fi rst committed (enzymatic) step in triglyceride (TG) biosynthesis, and thus, miR-27b upregulation in a short-term high-fat diet is likely an adaptive response to mitigate TG production and attenuate hyperlipidemia and hepatosteatosis. Likewise, elevated hepatic miR-27b also represses ANGPTL3 , which is produced and secreted from the liver to inhibit plasma TG lipolysis ( 92,93 ). Heparin sulfate proteoglycans on hepatocyte membranes contribute to the cellular uptake of TG-rich lipoproteins, and loss of NDST1, the heparin sulfate biosynthetic enzyme, results in decreased TG-rich and cholesterol-rich lipoproteins in the liver ( 88 ). Although diet-induced dyslipidemia results in elevated plasma cholesterol and TG, miR-27b upregulation in the liver likely protects the liver from extensive lipid accumulation by repressing NDST1 activity and heparin sulfate biosynthesis and, thus, reducing TG-rich lipoprotein uptake in the liver. Subsequent in vivo experimentation in Apoe Ϫ / Ϫ mice showed that in response to a high-fat/high-cholesterol diet, miR-27b is also signifi cantly gene regulation may also be highly relevant to the control of hepatic lipid metabolism (40)(41)(42)(43)(44)(45)(46)(47)(48)(49)(50). miRNAs have emerged not only as stable plasma biomarkers of metabolic states but also as attractive therapeutic targets for various cardiometabolic disorders ( 46,(51)(52)(53)(54)(55) ( 61,69 ). Although each of these classes of smRNA is thought to regulate gene expression through diverse mechanisms, miRNAs are currently the most widely studied and are known to be involved in posttranscriptional control of lipid homeostasis ( 56,57,70 ).
The human genome encodes for over 1,000 miRNAs, either within host protein-coding genes or as independent transcription units ( 56,71 ). Canonical biogenesis of a miRNA commences with RNA Polymerase II-mediated transcription, yielding a primary miRNA transcript (pri-miRNA) of variable length depending on the locus ( 56,59,72 ). The pri-miRNA is cleaved by a nuclear RNase complex DROSHA/DGCR8, generating a precursor miRNA (pre-miRNA) that has a hairpin-like secondary structure. The pre-miRNA is then exported via the exportin 5-dependent pathway to the cytoplasm, where it is subject to further enzymatic processing by DICER and its cofactors, producing a short dsRNA duplex ( ‫ف‬ 22 bp) ( 56,57,59,72 ). One strand of the duplex (mature miRNA) is loaded onto the RNA-induced silencing complex (RISC), which it guides and tethers to target mRNAs in order to regulate their stability and/or translational effi cacy ( 57,73 ). Some miRNAs are generated by alternative mechanisms, but they may still play a signifi cant role in lipid homeostasis. For example, certain miRNA clusters that are interspersed among Alu repeats in the genome have been shown to be transcribed by RNA Polymerase III ( 74 ). Also, short-length introns can form hairpin-like structures upon nascent transcription and be processed by RNA splicing machinery, rather than by DROSHA, and cleaved into pre-miRNA products. This class of miRNAs is referred to as miRtrons, due to their unique biogenesis from intronexon junctions (75)(76)(77)(78). Although very few miRtrons are extensively conserved, they are present in both invertebrates and mammals, suggesting that they may have evolved that isomiRs are loaded onto the RISC and recognize target mRNAs for gene repression ( 61,113,114 ). Some isomiRs actually match the genomic template, but differ in their 5 ′ -start and/or 3 ′ -end positions ( Fig. 1 ). This is likely due to variable enzymatic processing during biogenesis and/ or exonuclease-mediated nucleotide trimming. Changes to the 5 ′ -start position of a miRNA are expected to alter the canonical seed sequences, thereby reshuffl ing the profi le of target mRNAs ( Fig. 2 ). Recent studies have used gene reporter (luciferase) assays to demonstrate that two seed-shifted isomiRs have different targeting effects relative to their canonical counterparts ( 115,116 ). Some isomiR sequences may diverge from the genome due to posttranscriptional enzymatic editing and/or tailing of specifi c nucleotides ( Figs. 1, 2 ). The most dominant type of RNA editing is adenosine deamination, which is commonly referred to as adenosine-to-inosine (A-to-I) editing, and occurs with highest frequency in noncoding regions (e.g., regulatory RNAs such as pre-miRNAs, 3 ′ untranslated regions, etc.) ( 117,118 ). RNA editing events in the 5 ′ -end region of miRNA have been shown to redirect miRNA targeting activity and function ( 119,120 ). For example, A-to-I editing in the seed region of miR-376a results in a new pool of mRNA targets for miR-376a, which includes a new target (enzyme) in the uric acid metabolism pathway ( 121 ).
Posttranscriptional tailing, also referred to as nontemplated nucleotide addition (NTA), is carried out by a diverse set of nucleotidyl transferases ( Fig. 2 ). At least 7 out of the 12 known nucleotidyl transferases are thought to mediate 3 ′ NTA ( 19 ). For example, the PAP-associated domain family (PAPD4/GLD2, PAPD5) and the mitochondrial poly(A) polymerase (MTPAP) have been found to add adenines to the 3 ′ -end of miRNAs, including several that are highly relevant to hepatic function, such as miR-122 upregulated, while its target genes (GPAM and ANGPTL3) are signifi cantly downregulated ( 85 ). miR-27b overexpression in hepatocytes resulted in the signifi cant downregulation of 27 lipid-related genes ( 85 ). These fi ndings lead to the hypothesis that miR-27b plays an adaptive role in controlling systemic lipid levels.
Although the discovery of alternative modes of miRNA biogenesis is relatively recent, several noncanonical miRNAs have already been shown to be relevant to lipid metabolism. For example, a human miRtron, miR-1224, was found to be induced by lipopolysaccahride (LPS) and to directly target and regulate Sp1 transcription factor activity, which controls the expression of many lipid-related genes ( 77,104 ) including the LDL receptor (105)(106)(107). Another recent study found that miR-451, which bypasses DICER processing, is signifi cantly reduced in the livers of rats on high-fat diets ( 108 ), supporting an earlier study that found hepatic miR-451 to be downregulated in humans with nonalcoholic steatohepatitis (NASH) ( 109 ). The relevant targets of miR-451 remain unknown and merit further investigation. Also, two miRNAs generated from the human mitochondrial (mt) genome (miR-1974 and miR-1978), most likely smRNA fragments derived from mt tRNAs, were found to be signifi cantly upregulated in human coronary artery endothelial cells (HCAEC) treated with oxidized low-density lipoprotein (oxLDL) ( 110 ). Most interestingly, putative mRNA targets for these miRNA-like small RNAs are signifi cantly downregulated upon oxLDL treatment in HCAECs ( 110 ).

ISOMIRS OF CANONICAL MIRNAS
Irrespective of the biogenesis mechanism, a single genomic locus can give rise to multiple distinct isoforms (isomiR; Fig. 1 ) (111)(112)(113). Initially, isomiRs were either undetected by conventional profi ling methods, such as qPCR, or dismissed as experimental artifacts with limited biological relevance. Nevertheless, sophisticated new technologies and computational analyses of smRNA sequencing datasets have demonstrated that they are often present in far higher frequencies than previously appreciated. Moreover, subsequent biochemical experiments have revealed Fig. 1. Sources of isomiR diversity. The top panel illustrates sources of isomiR diversity, which stratify into two classes: templated and nontemplated variations. As illustrated by the bottom panel, an isomiR may contain one or both types of variations, and both templated (e.g., 5 ′ -shifts) and nontemplated (e.g., RNA edits) variations can create an isomiR with an altered seed. The seed region of each isomiR is underlined in blue.
PCR and NanoString technology ( 125 ). This high-throughput technique directly detects miRNA sequences, provides high precision, and allows for linear range of quantifi cation of a large number of samples ( 125 ). Another recent study has used mass spectrometry (MS) to observe and characterize miRNA isomiRs ( 122 ).
Several studies have reported that 3 ′ NTAs alter miRNA stability and activity (34)(35)(36)(37)(38)(39)(40). As such, NTAs represent a means by which miRNAs themselves are posttranscriptionally regulated, which adds another layer of biological complexity to miRNA-mediated gene regulation. The most prevalent 3 ′ NTAs are adenylation and uridylation, which appear to confer different functions for different miRNAs ( 31 ). One seminal study reported that the same miRNA can have varying 3 ′ -NTA frequencies across different developmental stages, indicating for the fi rst time that isomiR production is biologically regulated. For a detailed discussion of the biogenesis and functional signifi cance of isomiRs, we direct the reader to two comprehensive review articles (Refs. 112 and 126 ).
Currently, multiple strategies and platforms are available to detect and quantify miRNA isomiRs. A comprehensive and exhaustive review on miRNA profi ling, methods, and approaches is found in Ref. 123 . The most widely used technique for the quantifi cation of canonical miRNAs is real-time PCR using individual assays or low-density arrays. However, this method is not well-suited for isomiR quantifi cation, as reverse transcription primers and probes are not designed to detect 3 ′ -end variability. A signifi cant advantage of an alternate approach, high-throughput smRNA sequencing, is the ability to precisely resolve the 5 ′ -and 3 ′ -ends of miRNA sequences ( 124 ). Massive parallel sequencing, often referred to as high-throughput sequencing, takes advantage of amplifi ed clusters of small length cDNA prepared from small RNAs or DNA fragments, which are then simultaneously sequenced by imaging. smRNA sequencing can be labor intensive and time-inefficient, and it also may not provide a means of linear quantifi cation ( 125 ). Despite these limitations, the smRNA sequencing method is currently the most sensitive and, as such, has been the method of choice for isomiR analyses. A new, promising approach for isomiR analysis is digital NTAs have been demonstrated to alter AGO2-RISC loading; therefore, some 3 ′ NTAs may also indirectly alter miRNA function. Here we describe for the fi rst time isomiR changes associated with high-fat diets or dyslipidemia; are frequent, diverse, and sensitive to metabolic conditions, literature currently purports that 3 ′ -end NTAs do not mediate changes in direct miRNA targeting but, rather, alter miRNA stability and/or expression. Nevertheless, 3 ′ deletion of miR-122 in mice (Mir122ko) was found to cause severe steatosis as a result of the signifi cant accumulation of TG, but not cholesterol, in the liver ( 138 ). Specifi cally, de novo TG synthesis was found to be signifi cantly elevated while TG secretion was signifi cantly decreased in Mir122ko mice ( 138 ). Loss of liver miR-122 also resulted in decreased plasma total cholesterol, HDL-cholesterol, and LDL-cholesterol, but only a slight drop in plasma TG was reported ( 138 ). Most interestingly, 3 ′ -end adenylation by GLD-2 was found to stabilize miR-122, as GLD-2-null mice were reported to have increased levels of 3 ′ -uridylated miR-122 (degradation signal), decreased levels of mature miR-122, and increased levels of miR-122 target genes ( 122 ). Although GLD-2 is quite possibly involved in the 3 ′ -adenylation of other miRNAs, none other than miR-122 was reported to have signifi cantly reduced 3 ′ -adenylation however, the mechanism responsible for miRNA variability in the context of dyslipidemia is currently unknown. To our knowledge, none of the known NTA enzymes has been reported to be altered with high-fat diets or hyperlipidemia.
Canonical miR-122 and its isomiRs account for over 80% of miRNAs present in the liver. As such, miR-122 is the most studied hepatic miRNA and is directly involved in hepatitis C infection ( 127,128 ), cellular proliferation ( 129,130 ), differentiation ( 102 ), and tumor suppression ( 131 ). miR-122 is also a critical regulator of hepatic lipid metabolism and systemic lipid homeostasis ( 40,(132)(133)(134). miR-122 is thought to be transcribed only in hepatocytes; however, miR-122 is present in extracellular fl uids ( 135,136 ), including plasma, and may be transferred to other cells by extracellular RNA carriers ( 137 ). Liver-specifi c • Frequency of miRNA nontemplated nucleotide additions is not directly correlated with miRNA expression levels.
in GLD-2-null mice ( 122 ). Notably, 3 ′ -adenylation may have different effects on different miRNAs, as 3 ′ -adenylation of miR-26a failed to alter stability but, rather, signifi cantly reduced targeting capacity ( 139 ). This apparent discrepancy could be due differential nucleotide transfer enzymes, as miR-122 is poly-adenylated by GLD-2 and miR-26a was found to be poly-adenylated by ZCCHC11 ( 122,139 ). Although miR-122 is one of the most abundant miRNAs in the liver and one of the most highly modifi ed miRNAs, we found that overall the frequency of 3 ′ NTA is not correlated with miRNA expression ( Fig. 3D ).

SUMMARY
The study of miRNAs has provided insights into novel molecular mechanisms that underlie complex disease etiology. Accordingly, miRNAs have emerged as compelling drug targets for the treatment of numerous disorders, including cardiovascular disease and dyslipidemia. Recent technological advances, such as next-generation sequencing, small biologics, and integrative computational tools, have facilitated rapid progress in the profi ling and functional characterization of miRNAs. For example, there are now several cost-effi cient platforms and protocols for smRNA-seq, including those from Illumina, Life Technologies, and NanoString, that provide a high-resolution, digital representation of intracellular miRNA expression. Very recent computational analyses of smRNA-seq data have led to the discovery that many miRNAs are present in multiple isoforms, called isomiRs, indicating that the complete repertoire of functional miRNAs is likely more complex than previously appreciated. This review highlights the extensive diversity of miRNAs and the potential for complex posttranscriptional regulation of miRNAs as well as other noncoding smRNAs. In summary, miRNAs and other smRNAs control many facets of lipid metabolism, and this is not limited to the canonical forms but also likely includes related functional variants such as isomiRs.

PERSPECTIVES
• A diverse group of miRNAs and miRNA-like smRNAs control lipid metabolism.
• miR-27b is a posttranscriptional regulatory hub for lipid metabolism and regulates 27 genes related to lipid homeostasis.
• Many miRNAs previously reported to target lipidregulating genes are present in multiple isoforms, due to various types of 5 ′ and 3 ′ posttranscriptional modifi cations.
• Multiple sequencing and nonsequencing-based methods facilitate the quantifi cation and analysis of miRNA isomiRs.
• 5 ′ variability likely alters miRNA targeting activity and biological function.
• 3 ′ nontemplated nucleotide additions are miRNA specifi c and alter miRNA stability and/or loading onto the RISC.