Potential conflict of interest: Nothing to report.
Supported by the intramural programs of the National Human Genome Research Institute (NHGRI) and the National Heart, Lung, and Blood Institute (NHLBI), a K22 grant HL113039-01 (to K.C.V.) from the NHLBI, and an R00 grant 4R00DK091318-02 (to P.S.) from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK).
See Editorial on Page 432
Cellular and plasma lipid levels are tightly controlled by complex gene regulatory mechanisms. Elevated plasma lipid content, or hyperlipidemia, is a significant risk factor for cardiovascular morbidity and mortality. MicroRNAs (miRNAs) are posttranscriptional regulators of gene expression and have emerged as important modulators of lipid homeostasis, but the extent of their role has not been systematically investigated. In this study we performed high-throughput small RNA sequencing and detected ≈150 miRNAs in mouse liver. We then employed an unbiased, in silico strategy to identify miRNA regulatory hubs in lipid metabolism, and miR-27b was identified as the strongest such hub in human and mouse liver. In addition, hepatic miR-27b levels were determined to be sensitive to plasma hyperlipidemia, as evidenced by its ≈3-fold up-regulation in the liver of mice on a high-fat diet (42% calories from fat). Further, we showed in a human hepatocyte cell line (Huh7) that miR-27b regulates the expression (messenger RNA [mRNA] and protein) of several key lipid-metabolism genes, including Angptl3 and Gpam. Finally, we demonstrated that hepatic miR-27b and its target genes are inversely altered in a mouse model of dyslipidemia and atherosclerosis. Conclusion: miR-27b is responsive to lipid levels and controls multiple genes critical to dyslipidemia. (HEPATOLOGY 2013)
Cellular and plasma lipid levels are tightly controlled by complex feed-back and feed-forward mechanisms, which regulate the expression and activity of key metabolic genes1 at both the transcriptional and posttranscriptional levels.2, 3 Dysregulation of lipid metabolism can lead to hyperlipidemia, a major risk factor for cardiovascular disease.4 Several key processes for regulating cellular and systemic lipid levels have been identified5; however, posttranscriptional mechanisms remain less well characterized.
MicroRNAs (miRNAs) are short (≈22 nucleotides) noncoding RNAs that regulate gene expression at the posttranscriptional level.6, 7 They serve as stable plasma biomarkers for various disorders,8 are important factors in the pathogenesis of several diseases,9, 10 and are promising targets of novel therapeutic strategies.11, 12 In regard to lipid metabolic control, miRNAs have recently been found to modulate cholesterol homeostasis.13In vivo inhibition of a liver-specific miRNA, miR-122, significantly lowers plasma cholesterol levels in both mice and nonhuman primates.14-16 In addition, miR-33, which is encoded within an intron of SREBF2, is cotranscribed with SREBF2 and regulates the expression of the ATP-binding cassette transfer protein (ABCA1), a critical player in reverse cholesterol transport and in lipoprotein biogenesis.17-19 miR-33 has also been shown to regulate fatty acid oxidation in hepatic cell lines.20 Nevertheless, despite these important advances, the full extent of posttranscriptional control of lipid metabolism by miRNAs remains incompletely understood and has not been systematically investigated.21
Using an unbiased in silico approach, which should be generally applicable toward the identification of key regulatory miRNAs in any biological process, we predicted miR-27b as a regulatory hub in lipid metabolism. Furthermore, we demonstrated that hepatic miR-27b is responsive to lipid levels and regulates the expression (messenger RNA [mRNA] and protein) of key metabolic genes, including angiopoietin-like 3 (ANGPTL3) and glycerol-3-phosphate acyltransferase 1 (GPAM), which have been implicated previously in the pathobiology of lipid-related disorders.
Eight-week-old wildtype C57BL/6J mice were placed on either normal chow diet (4% fat, NIH-31 open chow, Zeigler Brothers, Gardners, PA) or a high-fat Western diet (21% fat, 42% calories from fat, ad libitum, TD88137, Harlan-Teklad, Frederick, MD) for 3 weeks (19-21 days). Adult (8-10 weeks) female apolipoprotein E null mice (Apoe−/−, C57BL/6J background; Jackson Laboratory, Bar Harbor, ME) were placed on either normal chow or cocoa butter diet with sodium cholate (16% fat, 37% calories from fat, 1.25% cholesterol, 0.125% choline chloride, 0.5% sodium cholate, TD90221, Harlan-Teklad) for 4 weeks (28 days). Mouse livers were excised and homogenized (100 mg) in Qiazol Total RNA extraction buffer. All mice were housed and the relevant studies were completed under active protocols approved by the National Institutes of Health, National Heart, Lung, and Blood Institute Animal Care and Use Committee. All protocols complied with, and all animals received humane care according to, the criteria outlined in the NIH “Guide for the Care and Use of Laboratory Animals.”
High-Throughput Small RNA Sequencing.
miRNA isolation and Illumina sequencing were completed as reported.22 Details are provided in the Supporting Methods.
Target sites (seed, centered) were predicted for miR-27b in both the 3′ untranslated regions (UTRs) and the open reading frames of the 151 lipid metabolism genes. Details of target site prediction and the identification of candidate miRNA regulatory hubs by Monte Carlo simulations are provided in the Supporting Methods.
Human hepatocytes (Huh7) were cultured in F12 Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% fetal bovine serum, penicillin (100 U/mL), and streptomycin (100 μg/mL), and maintained at 37°C with 5% CO2. For transfection, Huh7 cells were plated at 1 × 105 cells/mL for 24 hours prior to transient transfection (DharmaFECT4, Dharmacon, Lafayette, CO) with 10 nM miR-27b miRIDIAN mimic (C-300589-05) or 10/100 nM HI-27b miRIDIAN hairpin inhibitor (IH-300589-07) for 48 hours.
Gene Expression Studies.
Gene expression in Huh7 cells was assayed by microarray (Affymetrix GeneChIP Human Genome U133A 2.0), real-time polymerase chain reaction (PCR) (Applied Biosystems TaqMan assays), and enzyme-linked immunosorbent assay (ELISA) according to standard protocols. Details of the methodology and reagents used are provided in the Supporting Methods.
Reporter Gene (Luciferase) Assays.
Reporter gene assays were conducted with a truncated 3′ UTR (bases 1-806) of human GPAM (NM_020918.3), which was cloned downstream of firefly luciferase in a pEZX-MT01 vector (GeneCopoeia). Site-directed mutagenesis (QuickChange II XL, Stratagene), using custom primers (Supporting Table S4), was performed to alter the predicted miR-27b target site at base 329 (G>A). Transformation, DNA extraction, transient transfections, and Luciferase activity measurements were conducted according to standard protocols, which are described in detail in the Supporting Methods.
Murine plasma and hepatic lipid levels were measured according to standard enzymatic quantification (Roche Diagnostics). Details of blood collection, tissue extraction, and reagents used are provided in Supporting Methods.
When comparing two groups, Mann-Whitney nonparametric tests (two-tailed) were used unless otherwise stated. For all tests, P ≤ 0.05 was considered significant. All results are expressed as means ± standard error, and n = derives from independent experiments.
High-Throughput Small RNA Sequencing Detects at Least 150 MiRNAs in Mouse Liver.
To characterize mouse liver miRNAs, we performed high-throughput sequencing on a small RNA library generated from mouse liver and obtained ≈9.9 million small RNA reads (Materials and Methods). Using an in-house bioinformatic strategy, we determined that ≈40% (≈3.9 million) of the reads matched exactly (no mismatches) to 160 annotated mouse miRNAs in miRBase. Almost all of these miRNAs (n = 157) were represented by ≥3 exactly matching sequence reads and were thus identified as hepatic miRNAs (Fig. 1A; Supporting Table S1). The diversity and number of hepatic miRNAs is consistent with results from the few other previously published small RNA sequencing studies performed in other murine tissues.23, 24 The most highly abundant miRNA, miR-122, accounts for ≈90% of the miRNA-related sequence reads in the mouse liver (Fig. 1A). Nevertheless, many of the less abundant miRNAs have been shown to regulate important processes in the liver, such as miR-33 (cholesterol homeostasis, fatty acid oxidation),20, 25 miR-22 (hepatocyte proliferation),26 miR-125a-5p (lipid uptake),27 miR-30 (hepatobiliary development),28 and miR-29b (liver fibrosis).29
MiR-27b Is a Regulatory Hub in Lipid Metabolism.
A posttranscriptional “miRNA hub” in lipid metabolism was defined as an miRNA that is predicted to target more lipid metabolism-associated genes than expected by chance.30 To identify lipid metabolic miRNA hubs, we assembled a high-confidence list of 151 known lipid metabolism associated genes (Supporting Table S2) from three high-throughput screens: (1) a large-scale hepatic gene expression analysis (microarray) of transgenic mice overexpressing SREBF1 or SREBF231; (2) a systematic siRNA screen for lipid-regulating genes assayed by quantitative analysis of cellular cholesterol levels and LDL uptake32; and (3) a genome-wide screen for common genetic variants associated with plasma lipid levels.33
Typically, the most effective miRNA target sites occur within 3′ UTRs of mRNAs and have perfect base pairing with the “seed” region of the miRNA (nucleotides 2 through 7 from the 5′-end of the miRNA).34 For each of the 157 hepatic miRNAs identified by small RNA sequencing, we scanned the 3′ UTRs of the 151 known lipid metabolism-associated genes for seed-based target sites and the number of genes with at least one such predicted site was scored. We then performed Monte-Carlo simulations to obtain the expected number of genes predicted to be targeted by chance for each miRNA. Target sites for three hepatic miRNAs, namely, miR-27b, miR-128, and miR-365, were significantly overrepresented (empirical uncorrected P < 0.01) in the 151 known lipid metabolism genes. Among all mouse liver miRNAs, miR-27b had the most predicted lipid metabolism gene targets (n = 27) (empirical uncorrected P = 0.003) (Fig. 1B). We repeated this analysis for those miRNAs previously detected in human liver tissue by small RNA cloning,35 and again miR-27 was the most significant (Fig. 1C).
MiR-27b Levels Are Sensitive to Plasma and Hepatic Lipid Content.
To identify lipid-responsive hepatic miRNAs, we used high-throughput small RNA sequencing to quantify and compare miRNA expression in the livers of C57BL/6J mice on a normal chow diet and on a high-fat “Western” diet (HFD, 42% calories from fat). After 3 weeks, triglyceride levels (mg/dL) were significantly increased in the plasma (1.86-fold, P = 0.0006) (Fig. 2A) and liver (1.87-fold, P = 0.01) (Fig. 2B) of HFD mice compared to mice fed a normal chow diet. Analysis of the small RNA sequence data revealed that at least 50 miRNAs were ≥2-fold more abundant (percent of total reads) in HFD mouse liver (Fig. 2C; Supporting Table S1), including miR-27b, which was up-regulated 3.2-fold (Fig. 2D). To confirm this observation, we performed real-time PCR using individual TaqMan assays and found miR-27b to be significantly increased (2.4-fold, P = 0.03) in HFD livers compared to normal mouse livers (Fig. 2E). However, interestingly, levels of the primary transcript of miR-27b (pri-miR-27b) were not increased (Supporting Fig. S1).
MiR-27b Represses Critical Regulators of Lipid Homeostasis.
To validate miR-27b targeting of lipid metabolism genes experimentally, we transfected miR-27b in human hepatocytes (Huh7 cells) and performed whole-genome microarray expression analysis. Of the 13,785 unique genes assayed on the array, 1,318 were down-regulated at false-discovery rate (FDR) <0.05, including ≈10% of the original set of 151 known lipid metabolism genes. Of these 1,318 genes, 173 were down-regulated by a fold-change < −1.5 (Fig. 3A; Supporting Table S3). We applied the Sylamer algorithm36 to calculate enrichment scores for all possible 6-mer motifs in the 3′ UTRs of the 13,785 genes, sorted according to their level of down-regulation upon miR-27b overexpression. The most enriched 6-mer motif among down-regulated genes is CTGTGA (Fig. 3B), which is the exact reverse complement of the miR-27b “canonical seed” region (nucleotides 2-7 from the 5′-end of the miRNA).34 The two next-most enriched motifs, TGTGAA and ACTGTG, are reverse complements of the miR-27b seed region shifted by 1 nucleotide on either side (nucleotides 1-6 and 3-8 from the 5′-end of miR-27b, respectively). We next implemented an algorithm that identifies “seed” target sites. Among the significantly down-regulated genes with annotated 3′ UTR sequence (n = 161), ≈73% (n = 118) have canonical miR-27b seed sites, which represents a 2.2-fold enrichment compared to background expectation (Fisher's exact test, P < 0.0001) (Fig. 3C). Noncanonical “shifted seed” sites are present in an additional ≈9% (n = 14) of the down-regulated genes, leaving ≈18% (n = 29) of the genes without any predicted 3′ UTR target site.
To further validate miR-27b-mediated regulation of lipid metabolism genes, we introduced miR-27b mimics or inhibitors (antagomiRs) into Huh7 cells by transient transfection. Overexpression of miR-27b mimics resulted in a significant increase (552-fold, P = 0.02) in intracellular miR-27b levels, and inhibition of endogenous miR-27b resulted in a significant decrease (71% loss, P = 0.02) in intracellular miR-27b levels (Fig. 4A). We then assayed by real-time quantitative PCR the mRNA levels of six genes: Peroxisome proliferator-activated receptor gamma (PPARG), Angiopoietin-like 3 (ANGPTL3), N-deacetylase/N-sulfotransferase 1 (NDST1), 3-hydroxy-3-methylglutaryl-CoA reductase (HMGCR), Glycerol-3-phosphate acyltransferase 1, mitochondrial (GPAM), and Sterol regulatory element binding factor 1 (SREBF1). These six genes were selected on the basis of their well-established relevance to lipid metabolism (Supporting Methods).
Four of the six genes were significantly down-regulated by miR-27b overexpression (PPARG, P = 0.0006; ANGPTL3, P < 0.0001; NDST1, P = 0.0008; and GPAM, P < 0.0001; Fig. 4B-E) and one (HMGCR, P = 0.06) was just outside of significance at the 5% threshold (Fig. 4F). Inhibition of endogenous miR-27b significantly up-regulated the same four genes (PPARG, P = 0.01; ANGPTL3, P < 0.0001; NDST1, P = 0.02; and GPAM, P = 0.004; Fig. 4B-E). SREBF1 was not affected by miR-27b overexpression (Fig. 4G).
To assess the effect of miR-27b on the protein levels of these key lipid metabolism genes, secreted (ANGPTL3) and cellular (GPAM) protein levels were quantified by ELISA. Inhibition of endogenous miR-27b by transient transfection of Huh7 cells with antagomiRs significantly (P = 0.002) increased secreted ANGPTL3 protein levels in the media after 48 hours (Fig. 5A). miR-27b inhibition also resulted in significantly (P = 0.04) increased cellular GPAM levels (Fig. 5B). To determine if miR-27b modulates PPARG transcriptional activity, we performed PPARG binding assays with nuclear extracts from transfected Huh7 cells (Supporting Methods). Inhibition of endogenous miR-27b resulted in a significant (P = 0.01) increase in PPARG binding to immobilized response elements (Supporting Fig. S2). It should be noted that, whereas overexpression of miR-27b significantly reduced (39% loss, P = 0.002) secreted ANGPTL3 levels (Fig. 5A) after 48 hours, cellular GPAM protein levels and PPARG transcriptional activity were not affected (Fig. 5B; Supporting Fig. S2). These observations are likely explained at least in part by the stability and temporal dynamics of each protein.
Next we searched for canonical 3′ UTR seed-based miR-27b target sites within each of the six genes (Materials and Methods). As expected, SREBF1, which did not change (mRNA level) in response to overexpression of either miR-27b mimic or its antagomiR, did not harbor any canonical miR-27b seed sites (Fig. 4F). Three out of the five genes that were repressed by miR-27b (PPARG, NDST1, and GPAM) contained one or more seed sites within their 3′ UTRs (Fig. 4A-E). GPAM harbors two highly conserved and one moderately conserved miR-27b target site within its 3′ UTR (Fig. 4). To determine if miR-27b directly targets GPAM through one of these predicted sites, we performed reporter gene (luciferase) assays. A portion of the GPAM 3′ UTR, containing one putative miR-27b site, was cloned downstream of firefly luciferase (Materials and Methods). Dual transfection with miR-27b in HEK293 cells significantly (P = 0.001) reduced firefly luciferase activity (Supporting Fig. S3). After site-directed mutagenesis to eliminate the putative miR-27b site (Materials and Methods), miR-27b failed to knock-down firefly luciferase activity, indicating that the site is directly involved in miR-27b mediated regulation of GPAM (Supporting Fig. S3).
ANGPTL3 was the only down-regulated gene that did not harbor any miR-27b seed sites in its 3′ UTR. To further investigate the observed strong miR-27b-mediated regulation of ANGPTL3 (Fig. 4C, 5A), we expanded the search to two recently discovered classes of target sites: (1) 3′ UTR centered sites and (2) open reading frame (ORF) sites (Materials and Methods). As its name implies, 3′ UTR centered sites base pair to the center of the miRNA sequence,37 as opposed to the 5′-end seed region. Functional ORF sites are typically preceded by a stretch of rare codons,38 which can cause ribosomal pausing,39 thereby allowing miRNA silencing complexes to form stable interactions with the target site without ribosomal interference. We developed and implemented computational strategies to predict 3′ UTR centered sites based on the strength of base pairing to the center of a miRNA, and ORF seed sites based on a metric that evaluates codon rarity in the preceding sequence (Materials and Methods). Of the six genes examined, only ANGPTL3 was predicted to have both a 3′ UTR centered site and a high-confidence ORF target site for miR-27b (Fig. 4C). Furthermore, no other hepatic miRNA besides miR-27 was predicted to have high-confidence ORF or 3′ UTR sites in ANGPTL3.
Hepatic MiR-27b Is Up-regulated in a Mouse Model of Dyslipidemia and Atherosclerosis.
We placed 8-week-old Apoe−/− female mice on a high-fat/high-cholesterol diet (21% fat, 7.5% cocoa butter), which has been shown to induce severe hypercholesterolemia and advanced atherosclerosis.40, 41 To confirm the expected physiologic effects of this diet, we measured plasma total cholesterol and triglyceride levels after 4 weeks. We observed a significant increase in plasma cholesterol levels (4.6-fold, unpaired t test, P < 0.001; Fig. 6A) and a significant decrease in plasma triglycerides (≈63% loss, unpaired t test, P = 0.003; Fig. 6B) in the Apoe−/− mice fed the atherogenic diet.
After 4 weeks on the atherogenic diet, levels of both mature miR-27b (1.58-fold, unpaired t test, P = 0.09) and pri-miR-27b (unpaired t test, P = 0.03) were increased in the liver; Fig. 6C,D). Based on this finding, we next assessed the hepatic expression of miR-27b target genes. Consistent with the in vitro results, mRNA levels of both Angptl3 (≈30% loss) and Gpam (≈22% loss) were reduced (Supporting Fig. S4); however, these observations were outside of statistical significance.
In this study we provide in silico, in vitro, and in vivo evidence that miR-27b is a strong candidate regulatory hub in lipid metabolism. Based on Monte-Carlo simulations, miR-27b was predicted to target significantly more lipid metabolism-associated genes than expected by chance and more than any other hepatic miRNA. Two of the other miRNAs predicted to be regulatory hubs in lipid metabolism (Fig. 1B,C), miR-365 and miR-125, have previously been shown to play roles in either adipocyte differentiation42 or in cellular lipid uptake,27 respectively, thus validating our approach.
High-throughput small RNA sequencing and real time quantitative PCR analysis revealed that miR-27b is ≈3-fold up-regulated in the livers of mice on a high-fat diet. MiR-27b is encoded with miR-23b and miR-24-1 in the same cistron on mouse chromosome 13. Small RNA sequencing results suggest that both miR-23b and miR-24 are also up-regulated in the liver of wildtype mice after a high-fat diet by ≈2.2-fold and ≈7.9-fold, respectively. However, we did not detect any change in the levels of their primary transcript, which suggests that: (1) posttranscriptional mechanisms are completely responsible for the observed increase in the mature miRNA levels, or (2) there is an increase in transcription of the miR-27b locus but also a concomitant increase in the rate of processing of the primary transcript (i.e., decreased pri-miR-27b stability). In contrast, Apoe−/− mice on an atherogenic diet were found to have increased hepatic levels of both mature miR-27b and pri-miR-27b. Although the latter finding is suggestive of enhanced transcription, several other recently reported posttranscriptional mechanisms of miRNA expression control, including pri-miRNA tertiary structure43 and nontemplated nucleotide additions,44 may also be involved and cannot be ruled out.
Microarray and real-time quantitative PCR based gene expression analyses in human hepatocytes confirmed robust miR-27b-mediated regulation of key lipid metabolism genes, including PPARG, GPAM, and ANGPTL3. Studies in rodents have revealed that both GPAM and ANGPTL3 regulate lipid metabolism.45, 46 Recent genome-wide association studies in human populations have added to these findings, by identifying genetic polymorphisms in both GPAM and ANGPTL3 that are significantly associated with plasma lipid levels.33
GPAM is present in a variety of tissues; however, it is most highly expressed in the liver. It is known to catalyze the first committed step in de novo triglyceride synthesis,47 and, more recently, has been implicated in regulating cholesterol.33 As such, overexpression of GPAM in mouse liver results in fatty liver, hepatic steatosis, and plasma hyperlipidemia.48 Our data show that hepatic Gpam mRNA levels are reduced in Apoe−/− mice on a 4-week atherogenic diet, concomitant with a decrease in plasma triglyceride levels and an increase in hepatic miR-27b expression.
ANGPTL3 is expressed by the liver49 and secreted into circulation,50 where it suppresses the activity of lipoprotein lipase51 and endothelial lipase,52 which regulate triglyceride and HDL-cholesterol levels, respectively. Plasma levels of ANGPTL3 correlate with various parameters of lipid/carbohydrate metabolism53 and atherosclerosis,54 and specific nonsense mutations in ANGPTL3 lead to hypolipidemia.55 Although several tissues may contribute to plasma ANGPTL3 levels, our data in this study reveal that hepatic Angptl3 levels are decreased in Apoe−/− mice on a 4-week atherogenic diet, concomitant with an increase in hepatic miR-27b expression. It is possible that Gpam and Angptl3 are repressed by miR-27b in the adaptive response to dyslipidemic conditions, in order to mitigate the accumulation of lipids in circulation.
Further detailed in vivo experimentation is required to determine the extent to which miR-27b targeting of GPAM and ANGPTL3 is required for controlling plasma lipid levels, and whether modulation of endogenous miR-27b levels could serve as an effective therapeutic strategy for lipid-related disorders.
The authors thank Yanqin Yang, Ph.D. for help with bioinformatics, Alonzo Jalan for animal studies, Maureen Sampson for plasma lipid analysis, the NHLBI DNA Sequencing Core Facility (Jun Zhu, Ph.D.), the NHLBI Genomics Core Facility (Nalini Raghavachari, Ph.D.), and the NHGRI Microarray Core Facility (Abdel Elkahloun and Bhavesh Borate).