microRNA-146a-5p association with the cardiometabolic disease risk factor TMAO

Abstract

Trimethylamine-N-oxide (TMAO), a microbial choline metabolism byproduct that is processed in the liver and excreted into circulation, is associated with increased atherosclerotic lesion formation and cardiovascular disease risk. Genetic regulators of TMAO levels are largely unknown. In the present study, we used 288 mice from a genetically heterogeneous mouse population [Diversity Outbred (DO)] to determine hepatic microRNA associations with TMAO in the context of an atherogenic diet. We also validated findings in two additional animal models of atherosclerosis: liver-specific insulin receptor knockout mice fed a chow diet (LIRKO) and African green monkeys fed high-fat/high-cholesterol diet. Small RNA-sequencing analysis in DO mice, LIRKO mice, and African green monkeys identified only one hepatic microRNA (miR-146a-5p) that is aberrantly expressed across all three models. Moreover, miR-146a-5p levels are associated with circulating TMAO after atherogenic diet in each of these models. We also performed high-resolution genetic mapping and identified a novel quantitative trait locus on Chromosome 12 for TMAO levels. This interval includes two genes, Numb and Dlst, which are inversely correlated with both miR-146a and TMAO and are predicted targets of miR-146a. Both of these genes have been validated as direct targets of miR-146a, though in other cellular contexts. This is the first report to our knowledge of a link between miR-146 and TMAO. Our findings suggest that miR-146-5p, as well as one or more genes at the Chromosome 12 QTL (possibly Numb or Dlst), is strongly linked to TMAO levels and likely involved in the control of atherosclerosis.

INTRODUCTION

The metaorganismal pathway involving dietary choline, gut microbiota, liver enzymes, and trimethylamine-N-oxide (TMAO) has been implicated in cardiometabolic disease events in humans (6, 41). In a mouse model susceptible to atherosclerosis, dietary supplementation with choline (the dietary precursor of TMAO) or treatment with TMAO itself has been shown to promote atherosclerotic lesion development (42). TMAO production is thought to be dependent on the microbiota, as administration of antibiotics suppresses TMAO levels and atherosclerosis (23, 24). The centrality of gut microbiota in TMAO production and subsequent development of atherosclerosis is further demonstrated by adoptive transfer studies in mice (17).

Despite these advances, there is little known about the role of host genetics in contributing to variation in TMAO levels across the population. Although flavin monooxygenase 3 (FMO3) in the liver has emerged as an important regulator of TMAO production (4), human studies have shown that common genetic variation within the FMO3 gene is not associated with circulating TMAO levels (19). Genetic studies of inbred mouse strains have demonstrated that TMAO levels are heritable (33), and they have identified three candidate loci (3, 19) associated with circulating TMAO. These studies indicate that the genetic architecture and regulation of TMAO levels is complex and may extend beyond FMO3.

One possible mode of regulation of plasma TMAO levels is via microRNAs (miRNAs). Changes in gene and miRNA expression profiles have each been linked to the onset of many pathological conditions, including cardiovascular disease (CVD). miRNAs represent an important class of diet-responsive biomolecules that have emerged as prominent regulators of atherogenesis and related cardiometabolic conditions (13). For instance, several miRNAs, such as miR-33, miR-148a, miR-27, and others, have been implicated in such processes as lipoprotein metabolism, cholesterol synthesis, and lipid flux, all of which are directly related to atherosclerosis. However, there are no established connections between miRNAs and TMAO levels.

In this study, we utilize the Diversity Outbred (DO) mouse population, an ideal resource for systems genetics (9), as well as an additional mouse and monkey model of atherosclerosis, to investigate and uncover novel miRNA and genetic associations with circulating TMAO. Specifically, we report that only one miRNA, miR-146a-5p, is aberrantly elevated in the liver of all three models of atherosclerosis and significantly correlated with circulating TMAO. We also identify a novel link between a locus on Chromosome 12 and TMAO levels that is potentially mediated by interactions between miR-146 and two specific genes at this locus.

MATERIALS AND METHODS

DO mice.

Complete details on animal housing, husbandry, and handling, diet compositions, and methods of genotyping were previously reported in detail (38). In brief, 288 DO mice were maintained on a synthetic diet for 2 wk, fasted for 4 h, and then phenotyped for plasma clinical chemistries at 6 wk of age (Baseline). Following 2 wk of synthetic diet, mice were transferred to either a high-protein diet (HP) or an atherogenic diet [high-fat, high-cholesterol diet with added cholic acid (HFCA)]. Fasting plasma was taken after 18 wk on their respective diets. Phenotyping for plasma clinical chemistries after diet treatment (HFCA or HP) was performed on each sample. All procedures were approved by the Institution Animal Care and Use Committee (IACUC) at University of North Carolina Chapel Hill (IACUC protocol number 11-299).

Measurement of choline metabolites.

Measurement of TMAO in the DO mice was performed by the University of North Carolina Nutrition Obesity Research Center (Kannapolis, NC). Plasma was extracted with three volumes of acetonitrile spiked with internal standards of TMAO-d9 (DLM-4779-1, Cambridge Isotope Laboratories), incubated on ice for 10 min, and centrifuged at 15,000 g for 2 min. Quantification of TMAO was performed by liquid chromatography-stable isotope dilution-multiple reaction monitoring mass spectrometry. Chromatographic separations were performed on an Atlantis Silica HILIC 3 µm 4.6 × 150 mm column (Waters, Milford, MA) using a Waters ACQUITY UPLC system. The column was heated to 40°C, and the flow rate was maintained at 1 ml/min. The gradient was 5% A for 0.05 min, to 15% A in 0.35 min, to 20% A in 0.6 min, to 30% A in 1 min, to 45% A in 0.55 min, to 55% A in 0.05 min, at 55% A for 0.9 min, to 5% A in 0.05 min, at 5% A for 1.45 min, where A is 10% acetonitrile/90% water with 10 mM ammonium formate. The metabolites and their corresponding isotopes were monitored on a Waters TQ detector using characteristic precursor-product ion transitions: 76→58 for TMAO, 85→66 for TMAO-d9. Concentrations of TMAO in the samples were determined from its calibration curve using peak area ratio of the metabolite to its isotope. Samples from liver insulin receptor mice and African green monkeys were analyzed for TMAO by a previously reported method (43).

Measurement of miRNA and mRNA (gene) expression from DO mice.

Methods of RNA extraction from the livers of the DO mice, evaluation of RNA integrity, microarray analysis for messenger RNAs (mRNAs, which we refer to as genes for ease of reading), library prep, and small RNA sequencing were as previously published (10). miRNA expression was transformed by the Box-Cox method for expression quantitative trait loci (eQTL) mapping and correlation analyses. Differential expression analysis for miRNAs between HP (n = 129) and HFCA (n = 140) was performed with three methods, Student’s t-test of reads per million mapped to miRNAs (RPMMM) values, Student’s t-test of Box-Cox transformed RPMMM values, and DESeq. Correlations were calculated by the biweight midcorrelation method. Microarray and small RNA-Seq data are available on the Gene Expression Omnibus repository, accession number GSE99561.

QTL mapping.

QTL and eQTL mapping were performed with the DOQTL (15) package in R as previously described (38). Haplotype reconstructions were performed utilizing genotyping data from a much larger cohort of DO mice available at the Jackson Laboratory (kindly performed by Daniel Gatti, Jackson Laboratory). Diet was included as an additive covariate in the mapping model for measurements including those obtained from 24 wk old mice after dietary treatment. Significant QTL were determined at a genome-wide P value of <0.05, and suggestive QTL were determined at a P value of <0.63, which corresponds to one false positive per genome scan (26). QTL support intervals were defined by the 95% Bayesian credible interval, calculated by normalizing the area under the QTL curve on a given chromosome (36). The mapping statistic reported is log of the odds (LOD) ratio. The significance thresholds were determined via permutations of genome-wide scans by shuffling phenotypic or expression data in relation to individual genotypes. Association mapping was performed for TMAO levels or gene/miRNA expression with known single nucleotide polymorphisms (SNPs) within the identified QTL by imputing the founder SNPs onto the DO genomes. Candidate genes were identified by position based on the UCSC Genome Browser. To verify that QTL are robust, we repeated haplotype reconstruction and pre- and postdiet TMAO QTL mapping using the R/qtl2 package (http://kbroman.org/qtl2/), which differs from DOQTL in that it utilizes genotype calls instead of allele intensity plots for haplotype reconstruction.

TMAO measurements were transformed by the Box-Cox method to satisfy the model assumption of a normal distribution. Microarray robust multiarray average values were corrected for known SNPs in the DO founder strains as previously described (10). Each phenotype and miRNA were used to run 1,000 permutations. For gene-eQTL LOD thresholds, 500 genes were chosen at random to perform 1,000 permutations on each.

Bioinformatics.

MicroRNA target prediction in mouse reference (NCBI build 37) 3′-untranslated regions was performed with the TargetScan algorithm (2, 18). Positional conservation of a predicted target site in at least one other species was required. Correlation between microRNA levels and gene expression as well as TMAO was calculated by biweight midcorrelation analysis. All P values take into account multiple testing unless specifically noted otherwise. Significant differences between diet groups were determined by a Student’s t-test.

Small RNA sequencing in liver from liver-specific insulin receptor knockout mice.

Treatment and phenotyping of liver-specific insulin receptor knockout (LIRKO) mice have been previously reported by Miao et al. (29). All procedures were approved by the Institutional Animal Care and Research Advisory Committee at Boston Children's Hospital. Liver tissue was isolated from LIRKO mice (n = 4) and floxed controls (n = 4). RNA was extracted using Norgen Total RNA Purification Kit (Norgen Biotek, Thorold, ON, Canada). RNA yield was assessed by Thermo Scientific NanoDrop 2000 (Waltham, MA), and integrity was measured by Agilent 2100 Bioanalyzer (Santa Clara, CA). Small RNA library preparation using the TriLink CleanTag kit was performed at the Genome Sequencing Facility of Greehey Children’s Cancer Research Institute at the University of Texas Health Science Center at San Antonio and sequencing was carried out on the HiSeq platform at an average depth of ~50 million reads/sample (with a range across samples of 43 million to 52 million). Data processing and miRNA quantification were performed with miRquant 2.0 (22). On average, >80% of the trimmed reads mapped to the mouse genome, of which almost 60% corresponded to miRNA loci (Table 1).The length distribution of mapped reads revealed a clear peak at the size range 21–24 nucleotides, which matches the expectation for miRNAs. Differential expression analysis for miRNAs was performed by three methods, Student’s t-test of RPMMM values, Student’s t-test of Box-Cox transformed RPMMM values, and DESeq. The miR-146 result was consistent irrespective of the method used.

Table 1.

Mapping statistics for the small RNA-Seq performed on the LIRKO and floxed-control mice

Sample Name	Flox-1	Flox-2	Flox-3	Flox-4	LIRKO-1	LIRKO-2	LIRKO-3	LIRKO-4
Total reads	51973970	42902191	47850475	47869923	45908948	50369802	46750262	44212326
Trimmed reads	31165641	26910851	28363646	30454015	26775613	30214638	26949943	27703996
% Trimmed reads	59.96	62.73	59.28	63.62	58.32	59.99	57.65	62.66
Short reads	9172839	7832166	8383802	7791046	8478334	9375082	8032344	7561920
% Short	17.65	18.26	17.52	16.28	18.47	18.61	17.18	17.1
Exact match to genome	22727331	18989905	20622090	22038607	19005761	21649919	19207685	20742258
% EM	72.92	70.57	72.71	72.37	70.98	71.65	71.27	74.87
No exact match to genome	8438310	7920946	7741556	8415408	7769852	8564719	7742258	6961738
% NEM	27.08	29.43	27.29	27.63	29.02	28.35	28.73	25.13
Total mapped reads	25885900	21843204	24927900	24896302	22720147	25192463	23190437	24537737
% Mapped	83.06	81.17	87.89	81.75	84.85	83.38	86.05	88.57
Total mapped to miRs	14582406	8787063	16732984	12900178	13944230	15251346	14907174	17382853
% of total mapped to miRs	56.33	40.23	67.13	51.82	61.37	60.54	64.28	70.84
Total mapped to tRNAs	2156789	2615130	850218	2274703	1678002	1648429	1570099	1157747
% of total mapped to tRNAs	8.33	11.97	3.41	9.14	7.39	6.54	6.77	4.72

RNA-Seq, RNA sequencing; LIRKO, liver-specific insulin receptor knockout.

Small RNA sequencing in liver from monkeys.

Dietary treatment, liver tissue extraction, and phenotyping of African green monkeys has been previously reported by Chung et al. (8). The monkeys were housed in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited facility under the direct care of the Wake Forest School of Medicine Animal Resources Program. All experiments were approved by the Institutional Animal Care and Use Committees of Wake Forest School of Medicine (Institutional Animal Care and Use Committee protocol #A10-024). The samples used for reporting here were from monkeys fed either chow (n = 4) or a high-fat high-cholesterol (0.4 mg/kcal cholesterol) diet (n = 5). RNA was extracted using Norgen Total RNA Purification Kit (Norgen Biotek, Thorold, ON, Canada). RNA yield and quality were assessed by Thermo Scientific NanoDrop 2000 (Waltham, MA), and integrity was measured by Agilent 2100 Bioanalyzer (Santa Clara, CA). Small RNA library preparation using the TriLink CleanTag kit was performed at the Genome Sequencing Facility of Greehey Children’s Cancer Research Institute at the University of Texas Health Science Center at San Antonio, and sequencing was carried out on the HiSeq platform at an average depth of ~17 million reads/sample. Data processing and microRNA quantification were performed using miRquant 2.0 (22). On average, ~85% of the trimmed reads mapped to the human genome, of which ~65% corresponded to microRNA loci. The length distribution of mapped reads revealed a clear peak at the size range 21–24 nucleotides (Table 2), which matches the expectation for miRNAs. Differential expression analysis for miRNAs was performed using two methods, Student’s t-test and DESeq, based on RPMMM, Box-Cox transformed RPMMMs, and the same results were obtained from each method. Differential expression analysis for miRNAs was performed using three methods, Student’s t-test of RPMMM values, Student’s t-test of Box-Cox transformed RPMMM values, and DESeq. The miR-146 result was consistent irrespective of the method used.

Table 2.

Mapping statistics for the small RNA-Seq performed on the livers of the African green monkeys

Sample Name	Chow 1	Chow 2	Chow 3	Chow 4	HFHC 1	HFHC 2	HFHC 3	HFHC 4
Total reads	6921450	14800272	22431223	13731169	15695910	17844833	19020563	24498194
Trimmed reads	6578170	13894207	20261384	12891396	14732475	15751785	18132684	21045693
% Trimmed reads	95.04	93.88	90.33	93.88	93.86	88.27	95.33	85.91
Short reads	142744	572506	1590743	557777	613518	1659082	277830	2842095
% Short	2.06	3.87	7.09	4.06	3.91	9.30	1.46	11.60
Exact match to genome	3509573	9778874	13617582	8470357	10421206	11245175	11187318	13609384
% EM	53.35	70.38	67.21	65.71	70.74	71.39	61.70	64.67
No exact match to genome	3068597	4115333	6643802	4421039	4311269	4506610	6945366	7436309
% NEM	46.65	29.62	32.79	34.29	29.26	28.61	38.30	35.33
Total mapped reads	5117952.55	12539485.2	18139799.1	10992896.8	13102013.1	14518080	14570075.3	17626759.6
% Mapped	77.80	90.25	89.53	85.27	88.93	92.17	80.35	83.75
Total mapped to miRs	1997848.89	8934077.56	12645667.3	6966324.47	9266036.99	10660269.5	8962880.43	11253808.1
% of total mapped to miRs	39.04	71.25	69.71	63.37	70.72	73.43	61.52	63.85
Total mapped to tRNAs	532787.1	346637.4	717081	307478.7	670050.6	152697	1265460	954149.9
% of total mapped to tRNAs	10.41	2.76	3.95	2.80	5.11	1.05	8.69	5.41

HFHC, high-fat, high cholesterol diet.

RESULTS

Relatively few studies to date have reported host genetic associations with TMAO, and none have reported miRNA associations. In this study, we leverage a DO cohort of 288 female mice comprising 144 sibling pairs, which we have previously phenotyped for atherosclerosis and circulating lipids (10, 38), to identify novel genetic regulators of circulating TMAO. In the earlier studies, we reported baseline measures of metabolic parameters and molecular traits in the DO mice during feeding of a synthetic AIN-76 diet and then after feeding with either a high-fat, high-cholesterol diet with added cholic acid (HFCA) or a HP diet for 18 wk. In the current study, we measure plasma TMAO levels and integrate these data with hepatic mRNA (gene) expression and miRNA expression data from the same mice, to identify novel candidate miRNA and genetic regulators of TMAO concentrations in circulation.

Plasma TMAO levels in the DO mice and its relationship to cardiovascular risk factors.

At baseline the mean circulating TMAO value in the DO mice was 1.25 μM and ranged between 0.16 and 5.41 μM. There was no difference in TMAO levels between the mice randomized to each of the diet groups at baseline (data not shown). However, after the mice had been on an atherogenic diet for 18 wk, we observed a highly significant increase in TMAO levels, 4.19 ± 0.22 μM (P = 1.36 × 10⁻²²), whereas TMAO levels in the mice fed the HP diet did not differ significantly from the baseline levels (1.25 ± 0.06 μM) (Fig. 1A).

Fig. 1.

Plasma trimethylamine-N-oxide (TMAO) levels are altered by diet and are correlated with cardiometabolic phenotypes in the Diversity Outbred (DO) mice. A: mice were maintained on a synthetic diet for 2 wk, fasted for 4 h, and then phenotyped for plasma clinical chemistries at 6 wk of age (Baseline). Following 2 wk of synthetic diet, mice were transferred to either a high-protein diet (HP, green) or an atherogenic diet [high-fat/high-cholesterol (HFHC), purple]. Plasma was taken from 24 wk old mice after 18 wk on their respective diets and then phenotyped for plasma clinical chemistries after diet treatment (HFCA or HP). ****P < 0.0001. Correlation plots demonstrating the associations between postdiet TMAO and postdiet VLDL/LDL-C (B) or total cholesterol (C) or atherosclerotic lesion size (D). Lesion sizes were measured in oil red O-stained slides of cross sections of aortic sinuses from each mouse and averaged based on number of slides per mouse. Data are representative only of HFCA-fed mice, as HP-fed mice developed no lesions. Representative lesions from DO mice fed either HP (E) or HFCA diet (F).

We next assessed the relationship between TMAO and cardiometabolic risk factors. We found a significant correlation between TMAO and VLDL/LDL cholesterol (R = 0.60, P = 2.05 × 10⁻²⁵) (Fig. 1B) as well as total cholesterol (R = 0.54, P = 1.23 × 10⁻¹⁹) (Fig. 1C). A correlation analysis across samples from mice fed the HFCA diet revealed a moderate relationship between TMAO and atherosclerosis (R = 0.22, P = 0.03) similar to previous reports (3, 4, 38) (Fig. 1D), indicating that a relationship between TMAO and atherosclerosis is likely driven by a few of the more susceptible mice in the study.

Identification of robustly expressed miRNAs with significant correlations to TMAO after dietary treatment.

In a previous study, we reported the miRNA abundance in the livers of the DO mice and performed differential expression analysis on all robustly expressed miRNAs (10). From this analysis, we focused on those miRNAs with at least a 1.5-fold change in expression at a false discovery rate P < 0.05 in mice fed the atherogenic diet compared with those fed the control diet. Of these, we found that 24 miRNAs are significantly downregulated and 38 miRNAs upregulated by the atherogenic diet (Table 3).

Table 3.

Liver microRNAs correlated with circulating TMAO in DO mice

miRNAs	Mean of HFCA Samples	Mean of HP Samples	Fold Change	FDR for t-Test	Correlation to Postdiet TMAO	Correlation FDR
miR-34a-5p	129.20	21.63	5.97	1.33 × 10−29	0.62	9.46 × 10−25
miR-1247-5p	124.02	26.93	4.61	1.82 × 10−32	0.55	7.67 × 10−19
miR-146b-5p	231.73	54.04	4.29	8.77 × 10−15	0.40	8.30 × 10−10
miR-342-3p	156.55	45.60	3.43	3.35 × 10−14	0.43	6.42 × 10−11
miR-199a-1-3p	365.69	140.93	2.59	1.34 × 10−17	0.48	6.18 × 10−14
miR-199b-3p	365.69	140.93	2.59	1.34 × 10−17	0.48	6.18 × 10−14
miR-199a-2-3p	366.95	141.45	2.59	1.34 × 10−17	0.48	6.18 × 10−14
miR-142-5p_-_2	155.88	60.59	2.57	2.74 × 10−17	0.45	5.01 × 10−12
miR-21-5p_-_1	69.05	27.42	2.52	3.72 × 10−15	0.40	1.74 × 10−09
miR-200c-3p	122.57	51.86	2.36	4.36 × 10−08	0.31	4.99 × 10−06
miR-146a-5p	2335.51	1063.79	2.20	4.81 × 10−16	0.42	1.51 × 10−10
miR-214-3p_-_1	107.05	50.05	2.14	3.20 × 10−16	0.43	2.83 × 10−11
miR-200b-3p	616.52	292.09	2.11	1.73 × 10−06	0.22	2.57 × 10−03
miR-200a-3p	279.85	135.38	2.07	2.99 × 10−07	0.23	1.44 × 10−03
miR-15b-5p	80.59	39.72	2.03	9.98 × 10−16	0.37	2.37 × 10−08
miR-222-3p	408.62	201.93	2.02	1.69 × 10−28	0.52	1.22 × 10−16
miR-221-3p	1102.77	556.82	1.98	1.87 × 10−21	0.44	9.67 × 10−12
miR-214-3p	191.42	97.10	1.97	4.67 × 10−14	0.43	6.39 × 10−11
miR-151-5p	62.22	32.20	1.93	3.68 × 10−12	0.28	7.07 × 10−05
miR-199a-2-5p	138.79	74.40	1.87	2.63 × 10−14	0.41	2.72 × 10−10
miR-199a-1-5p	138.79	74.40	1.87	2.63 × 10−14	0.41	2.72 × 10−10
mmu-let-7e-5p	650.81	370.31	1.76	9.33 × 10−20	0.50	1.33 × 10−14
miR-150-5p	134.48	77.13	1.74	5.61 × 10−08	0.33	1.25 × 10−06
mmu-let-7i-5p	5787.62	3394.20	1.71	2.16 × 10−16	0.39	2.89 × 10−09
miR-1839-5p_+_1	113.62	67.17	1.69	9.81 × 10−26	0.46	1.37 × 10−12
miR-5117-3p_-_4	385.05	239.59	1.61	4.86 × 10−06	0.22	2.64 × 10−03
miR-23a-3p	537.36	337.43	1.59	1.82 × 10−09	0.32	2.96 × 10−06
miR-125b-1-5p_+_1	65.75	41.63	1.58	3.52 × 10−11	0.31	4.72 × 10−06
miR-21-5p	29835.80	18986.12	1.57	4.30 × 10−09	0.28	5.57 × 10−05
miR-125b-2-5p_+_1	65.08	41.60	1.56	9.61 × 10−11	0.31	5.20 × 10−06
miR-125b-1-5p	2530.96	1635.39	1.55	1.25 × 10−16	0.39	5.30 × 10−09
mmu-let-7a-1-5p	9296.91	6056.05	1.54	1.20 × 10−18	0.40	9.60 × 10−10
miR-125b-2-5p	2381.69	1551.99	1.53	9.71 × 10−16	0.38	1.21 × 10−08
miR-181b-1-5p	80.04	52.40	1.53	7.54 × 10−05	0.38	6.48 × 10−09
mmu-let-7a-2-5p	9094.37	5964.05	1.52	3.04 × 10−18	0.40	1.84 × 10−09
miR-125b-2-3p	69.34	45.71	1.52	8.53 × 10−06	0.17	2.54 × 10−02
miR-191-5p_+_1	50.19	33.14	1.51	4.22 × 10−17	0.35	1.80 × 10−07
miR-181b-2-5p	96.62	64.26	1.50	1.48 × 10−04	0.37	4.05 × 10−08
miR-92a-1-3p	11660.97	17700.75	−1.51	2.10 × 10−15	−0.29	3.30 × 10−05
miR-127-3p	76.23	117.77	−1.54	2.68 × 10−05	−0.23	1.14 × 10−03
miR-451	74.77	116.57	−1.56	3.27 × 10−06	−0.17	1.97 × 10−02
miR-671-5p	34.98	55.32	−1.59	9.57 × 10−13	−0.26	1.92 × 10−04
miR-320-3p_+_1	165.60	262.73	−1.59	1.34 × 10−17	−0.31	8.41 × 10−06
miR-192-5p	16031.71	25608.91	−1.59	1.82 × 10−32	−0.42	9.82 × 10−11
miR-5109_-_6	417.51	684.43	−1.64	2.67 × 10−03	−0.13	9.51 × 10−02
miR-5109_-_3	73.34	122.36	−1.67	6.52 × 10−06	−0.22	1.89 × 10−03
miR-5109_-_4	181.40	311.34	−1.72	7.35 × 10−05	−0.18	1.40 × 10−02
miR-5105_+_8	149.61	261.14	−1.75	2.46 × 10−07	−0.23	1.38 × 10−03
miR-320-3p	1829.28	3208.55	−1.75	3.18 × 10−20	−0.34	3.07 × 10−07
miR-122-3p	90.29	159.01	−1.75	6.62 × 10−15	−0.34	2.77 × 10−07
miR-5105_+_7	74.04	130.56	−1.75	3.32 × 10−12	−0.29	2.07 × 10−05
miR-1981-5p	312.76	554.51	−1.78	3.28 × 10−17	−0.36	1.11 × 10−07
miR-5109_-_5	58.97	105.26	−1.78	1.84 × 10−04	−0.17	2.47 × 10−02
miR-874-3p	135.20	254.03	−1.89	1.66 × 10−15	−0.29	2.49 × 10−05
miR-122-3p_+_1	67.57	132.65	−1.96	5.57 × 10−19	−0.38	8.85 × 10−09
miR-5115_-_6	134.11	264.68	−1.96	6.69 × 10−06	−0.08	3.88 × 10−01
miR-5109_+_7	33.54	66.90	−2.00	2.25 × 10−05	−0.25	3.74 × 10−04
miR-5115_-_7	59.31	118.45	−2.00	9.45 × 10−05	−0.03	7.35 × 10−01
miR-5115_-_3	29.89	60.55	−2.04	2.16 × 10−05	−0.04	6.33 × 10−01
miR-5115_-_5	42.02	94.58	−2.27	8.22 × 10−08	−0.20	4.98 × 10−03
miR-5115	47.53	116.22	−2.44	1.69 × 10−04	−0.11	1.96 × 10−01
miR-5115_-_8	88.08	690.92	−7.69	3.89 × 10−03	−0.11	1.73 × 10−01

Shown are the microRNAs (miRNA or miR) that are significantly differentially expressed by at least 1.5-fold in the Diversity Outbred (DO) mouse livers, and the correlations of each miRNA to postdiet trimethylamine-N-oxide (TMAO). FDR (false discovery rate) calculated by Benjamini-Hochberg method. Correlations calculated using biweight midcorrelation. HP, high-protein diet.

We next sought to determine if any of these robustly expressed and significantly altered miRNAs are associated with postdiet TMAO levels. We found that 58 of the miRNAs are significantly correlated with postdiet TMAO levels (Table 3).

Twelve miRNAs are significantly altered in both DO mice fed an atherogenic diet and in liver insulin receptor knockout mice.

The LIRKO mouse is a model of atherosclerosis and cardiometabolic dysfunction (5). LIRKO mice are known to have elevated TMAO levels in circulation compared with their floxed controls even when fed a chow diet (29), and our measurements of circulating TMAO corroborated this finding (Fig. 2A). When fed a high-cholesterol diet with added cholic acid, the mice become hypercholesterolemic and develop severe atherosclerosis (29). To determine which if any of the miRNAs most strongly correlated with TMAO in the DO mice are also significantly elevated in an independent model documented to have elevated circulating TMAO, we performed small RNA-Seq in liver tissue from both LIRKO (n = 4) and floxed control mice (n = 4) (Table 1). Quantification with the tool miRquant 2.0 (22) and differential expression analysis identified 44 miRNAs with a fold-change of at least 1.5 in the livers of LIRKO mice compared with floxed controls (Fig. 2B, Table 4). Twelve of these are also significantly altered in the livers of the DO mice fed an atherogenic diet. (Table 4).

Fig. 2.

microRNA (miR)-146a is the only coincident miRNA across three animal models. A: TMAO levels in the liver-specific insulin receptor knockout (LIRKO) mice (n = 4) relative to the floxed control mice (n = 4). B: fold-change in miRNA expression in LIRKO mouse livers. Red circles are miRNA that are upregulated, and blue indicate downregulated miRNA. C: circulating TMAO levels in African green monkeys fed high-fat/high-cholesterol (HFHC, n = 5) diet compared with baseline chow (n = 4). Error bars show standard error. D: fold-change in miRNA expression in the livers of the African green monkeys fed HFHC diet relative to chow-fed monkeys. Red circles are miRNA that are upregulated, and blue indicate downregulated miRNA. E: Venn diagram depicting that there is one coincident miRNA that is significantly upregulated at least 1.5-fold or more across all three models of cardiometabolic dysfunction and elevated TMAO, miR-146a. *P value < 0.05.

Table 4.

Significantly differentially expressed miRNAs in LIRKO mouse livers

miRNAs	Average Expression		Fold Change	P Value
	LIRKO	Floxed Controls
mmu-mir-34a-5p	1144.70	201.58	5.68	6.68 × 10−05
mmu-mir-802-5p	610.42	210.19	2.90	9.76 × 10−04
mmu-mir-150-5p	90.84	32.90	2.76	3.56 × 10−02
mmu-mir-802-3p	85.38	31.85	2.68	3.48 × 10−04
mmu-mir-125a-5p	1059.51	438.58	2.42	3.20 × 10−04
mmu-let-7e-5p	1722.41	825.38	2.09	8.29 × 10−04
mmu-mir-484	190.3175	93.58	2.03	4.18 × 10−06
mmu-mir-99b-5p	987.73	497.95	1.98	1.89 × 10−03
mmu-mir-674-5p	135.36	71.10	1.90	1.27 × 10−04
mmu-mir-132-3p	103.91	55.96	1.86	1.46 × 10−04
mmu-mir-15b-5p	102.32	55.77	1.83	2.87 × 10−03
mmu-mir-29a-3p_-_1	522.33	288.48	1.81	6.63 × 10−04
mmu-mir-29a-3p	5710.65	3157.46	1.81	2.65 × 10−04
mmu-mir-351-5p	89.86	50.27	1.79	7.12 × 10−04
mmu-mir-320-3p	526.49	295.98	1.78	1.29 × 10−02
mmu-mir-1981-5p	81.83	46.23	1.77	3.14 × 10−03
mmu-mir-378-5p	310.19	176.52	1.76	2.49 × 10−04
mmu-mir-29b-1-5p	48.76	27.81	1.75	5.60 × 10−03
mmu-mir-320-3p_+_1	43.18	24.75	1.74	1.44 × 10−02
mmu-mir-328-3p	196.27	115.04	1.71	1.61 × 10−02
mmu-mir-146a-5p	2797.98	1643.65	1.70	4.66 × 10−02
mmu-mir-106b-3p	128.50	75.70	1.70	9.91 × 10−06
mmu-mir-652-3p	142.14	83.79	1.70	4.21 × 10−04
mmu-mir-140-3p_+_1	881.15	519.98	1.69	1.22 × 10−04
mmu-mir-339-5p	101.31	61.38	1.65	1.31 × 10−03
mmu-mir-28-5p	831.35	516.81	1.61	1.85 × 10−02
mmu-mir-423-3p	575.67	360.44	1.60	1.26 × 10−02
mmu-mir-139-5p	309.94	196.50	1.58	2.00 × 10−04
mmu-mir-30c-1-3p	50.29	32.42	1.55	2.85 × 10−04
mmu-mir-29a-3p_+_6	78.18	52.03	1.50	3.77 × 10−03
mmu-mir-122-3p	132.34	201.89	−1.51	2.93 × 10−02
mmu-mir-203-3p	716.29	1109.83	−1.54	2.56 × 10−03
mmu-mir-122-3p_+_1	139.30	216.96	−1.56	3.59 × 10−02
mmu-mir-99a-5p	201.68	314.37	−1.56	1.18 × 10−02
mmu-mir-210-3p	51.41	82.12	−1.59	7.64 × 10−04
mmu-mir-5105_+_7	39.05	68.61	−1.75	3.89 × 10−03
mmu-mir-31-5p	908.00	1624.08	−1.79	1.79 × 10−02
mmu-mir-5105_+_8	117.25	222.41	−1.89	6.25 × 10−05
mmu-mir-29c-3p	434.60	834.13	−1.92	1.07 × 10−03
mmu-mir-1948-3p	28.19	56.56	−2.0	5.21 × 10−05
mmu-mir-182-5p	50.56	148.86	−2.94	3.13 × 10−03
mmu-mir-455-3p	114.85	395.03	−3.45	3.16 × 10−03
mmu-mir-455-5p	23.29	131.03	−5.55	1.88 × 10−04
mmu-mir-455-3p_+_1	6.91	40.36	−5.88	1.16 × 10−03

miRNAs that fit the criteria of having a fold change of at least 1.5 and a P value < 0.05 in the liver tissue of LIRKO mice compared with floxed controls. Boldfaced microRNAs are those in common with DO mice.

miR-146a-5p is the only aberrantly expressed miRNA shared in an independent monkey model of atherosclerosis.

We next sought to provide further validation of our results by investigating a third animal model of hypercholesterolemia and atherosclerosis. Specifically, we profiled miRNAs by small RNA-Seq in liver tissue from African green monkeys either during standard chow diet feeding (n = 4) or after a 10 wk high-fat/high-cholesterol (HFHC) atherogenic diet (n = 5) (Table 2). The monkeys fed the HFHC diet exhibited significantly elevated circulating TMAO levels (Fig. 2C). After differential expression analysis, we found that nine miRNAs were significantly altered by 1.5-fold or more (Fig. 2D, Table 5). Only one of these miRNAs, miR-146a-5p, is in common with the robustly altered miRNAs identified in the DO (Table 1) and LIRKO mice (Fig. 2E). A list of all miRNAs upregulated are shown in Supplemental Table S1. (The online version of this article contains supplemental material.) Notably, miR-146a-5p is almost twofold upregulated in the liver of both LIRKO mice and HFHC-fed monkeys (Fig. 3, A and B). We also found that the levels of miR-146a-5p are strongly correlated with plasma TMAO levels in both models (Fig. 3C), as well as with total plasma cholesterol, LDL cholesterol, and HDL cholesterol in HFHC-fed monkeys (Fig. 3, D–F).

Table 5.

Significantly differentially expressed microRNAs in African green monkey livers

miRNAs	Average Expression		Fold Change	P Value
	HFHC	Chow
hsa-mir-146a-5p	2237.55	1103.06	2.03	2.81 × 10−02
hsa-mir-21-5p	19102.80	9462.00	2.02	2.46 × 10−03
hsa-mir-126-5p	4219.41	2169.08	1.95	7.22 × 10−03
hsa-mir-126-3p_+_1	954.55	497.72	1.92	5.23 × 10−03
hsa-mir-22-3p	38512.74	20621.43	1.87	1.99 × 10−02
hsa-mir-660-5p	458.84	248.33	1.85	1.61 × 10−02
hsa-mir-126-3p	1966.37	1121.34	1.75	3.22 × 10−02
hsa-mir-451a	133.67	788.84	−5.88	4.82 × 10−03
hsa-mir-486-5p	1157.44	8391.34	−7.14	9.12 × 10−04

miRNAs that fit the criteria of having a fold change of at least 1.5 and a P value < 0.05 in HFHC-fed monkeys compared with chow-fed controls. The boldfaced microRNA is the only one in common with DO mice and LIRKO mice.

Fig. 3.

miR-146a is upregulated and increases with TMAO levels. A: miR-146a expression in LIRKO mice (n = 4) vs. floxed controls (n = 4). B: miR-146a expression in HFHC-fed African green monkeys (n = 5) vs. chow-fed controls (n = 4). C: correlation plot of miR-146a to TMAO levels in LIRKO and floxed mice (black) and in African green monkeys (gray) on HFHC or chow diet. D–F: correlation plots demonstrating correlations between hepatic miR-146a-5p expression and circulating total cholesterol levels (A), circulating LDL cholesterol (B), and circulating HDL cholesterol (C) in African green monkeys. R is the Pearson correlation coefficient. *P < 0.05.

We hypothesized that the underlying mechanism of the association between miR-146a-5p and TMAO could be: 1) a direct genetic connection shared between the two, and/or 2) genes targeted by miR-146a-5p that are directly or indirectly causing an increase in TMAO.

Genetic locus on Chromosome 12 is associated with circulating TMAO levels, but not miR-146a-5p expression.

To test our first hypothesis, we performed QTL mapping for TMAO and identified a novel QTL on Chromosome 12 (LOD = 7.04, P < 0.1) associated with TMAO after dietary treatment (Fig. 4, A and B). The QTL has a ~3.6 Mb support interval (83.46–87.08 Mb), which contains 80 genes (Supplemental Table S2). We note that this locus is robust as the QTL analysis of baseline plasma samples identified a similar region on Chr 12 (83.5–87.1 Mb) that is significantly associated with TMAO. miR-146a-5p is physically located at Chr 11: 43 Mb, thus eliminating it as a direct candidate at the locus. We next performed eQTL analysis for miR-146a-5p (Fig. 4C). While miR-146a-5p is significantly upregulated by an HFCA diet (Fig. 5A) and correlated with postdiet TMAO levels in the DO mice (Fig. 5B), we did not identify a genome-wide significant cis-eQTL signal, although there is a suggestive trans-eQTL (LOD = 6.22, P < 0.63) located between 57.3 and 63.3 Mb on Chromosome 18 (Supplemental Table S3). These findings suggest that our first hypothesis, direct regulation of TMAO by miR-146a-5p, is unlikely and that this association is mediated by another factor.

Fig. 4.

Quantitative trait locus (QTL) mapping for TMAO and miR-146a expression. Genome-wide QTL scan for loci affecting circulating postdiet TMAO levels (A) and miR-146a expression (C). For each plot, the red horizontal line denotes a genome-wide P value threshold of 0.1, and the gold line denotes a P = 0.63 (representing at least one false positive). Thresholds for TMAO and miR-146a were calculated by running permutations on the data sets. B: the eight coefficients of the QTL model show the effect of each founder allele on the phenotype along Chromosome 12. n = 288 in all panels.

Fig. 5.

miR-146a is associated with TMAO and its predicted target genes. A: expression of miR-146 in DO mice by diet group. ****P < 0.0001. B: hepatic miR-146a-5p expression and circulating TMAO levels. C: hepatic expression of Numb and circulating TMAO levels. D: hepatic expression of Dlst and circulating TMAO levels. E: hepatic miR-146a-5p expression and hepatic expression of Numb. F: hepatic miR-146a-5p expression and hepatic expression of Dlst in the DO mice. R is the biweight midcorrelation coefficient; P is the corrected P value (Benjamini-Hochberg method). miR-146a normalized RPMMM counts and postdiet TMAO values were transformed by the Box-Cox method. n = 288 in all panels.

Several genes in the QTL interval are significantly correlated with TMAO levels and are predicted and validated targets of miR-146a-5p.

To test our second hypothesis, we utilized TargetScan to identify predicted targets of miR-146a-5p (materials and methods). Of the 294 transcripts predicted to be targeted by miR-146a-5p, we found that six predicted targets are located within the TMAO QTL interval: Zfyve1, Psen1, Numb, Elmsan1, Dlst, and Gpatch2l (Table 6). Using microarray data for gene expression in the livers of the DO mice, we next performed a correlation analysis to determine if the expression of any of the six genes is associated with postdiet TMAO levels (Table 6). We found that only three, Dlst, Elmsan1 (data not shown), and Numb, are significantly inversely correlated with postdiet TMAO levels (−0.36, −0.23, and −0.25, respectively) (Fig. 5, C and D). Of these three, only Dlst and Numb are also significantly inversely correlated with miR-146a-5p expression (Table 6; Fig. 5, E and F), which is the expected direction of association between a miRNA and its target gene. Notably, the regulatory interaction between miR-146a-5p and Numb has been validated in several different tissues (14, 21), and the interaction between miR-146a-5p and Dlst has recently been validated in cardiomyocytes (20), though neither has been studied previously in the liver. Overall, these results point to a strong link between miR-146a/b and TMAO, with Numb and Dlst as potential mediators, particularly in the context of an atherogenic diet.

Table 6.

Predicted miR-146a-5p targets in the TMAO QTL region

Gene	Correlation to TMAO	TMAO Correlation P Value	Correlation to miR-146a-5p	miR-146a-5p Correlation P Value
Dlst	−0.36	4.27 × 10−08	−0.48	3.04 × 10−15
Elmsan1	−0.23	7.69 × 10−03	−0.15	5.42 × 10−02
Gpatch2l	−0.08	7.45 × 10−01	−0.018	9.21 × 10−01
Numb	−0.25	5.57 × 10−04	−0.38	4.14 × 10−09
Psen1	0.03	9.66 × 10−01	0.18	3.26 × 10−02
Zfyve1	−0.045	9.61 × 10−01	−0.22	1.72 × 10−02

TMAO quantitative trait locus (QTL) candidate genes that are predicted targets of miR-146a-5p. Correlations were calculated using the biweight midcorrelation. Boldfaced rows denote genes that are significantly inversely correlated with both TMAO and miR-146a-5p.

miR-146a-5p and TMAO are both correlated with inflammation driver NF-κB.

Inflammation is a well-known hallmark of atherosclerosis and metabolic diseases. Indeed, in the DO mice, we find that there is an increase in liver inflammation as denoted by an increase in circulating AST (aspartate transaminase) and ALT (alanine transaminase) levels (10). A well-established regulator of proinflammatory signals, NF-κB, is also known to promote miR-146a-5p expression (40). In the DO mice, Nfkb1 is significantly upregulated at the mRNA level (fold-change = 1.4, P = 1.75 × 10⁻³⁷) (Fig. 6A) and is very strongly correlated with miR-146a-5p expression (R = 0.68, P = 2.36 × 10⁻³⁵) (Fig. 6B). In addition, Nfkb1 expression is also significantly correlated with TMAO levels (R = 0.53, P = 6.76 × 10⁻¹⁸) (Fig. 6C).

Fig. 6.

NF-κB is associated with TMAO and miR-146a. A: hepatic expression of NF-κB. ****P < 0.0001. Correlation plots of NF-κB expression with miR-146a expression (B) and TMAO levels (C). Correlation plots of Cd74 expression with miR-146a expression (D) and postdiet TMAO levels (E). Correlation coefficients were calculated by the biweight midcorrelation method, and P values were adjusted by Benjamini and Hochberg method. n = 288 in all panels.

Although the miR-146a-5p eQTL is seemingly unrelated to the TMAO QTL, there are several genes located in the region, notably Cd74, which are related to inflammatory pathways involving NF-κB (Supplemental Table S2). Cd74 has been established as an activator of NF-κB (16, 39), which in turn has been shown to promote transcription of miR-146a-5p, suggesting that Cd74 may be an upstream indirect regulator of miR-146a-5p. In the DO mice, Cd74 mRNA expression in the liver is very strongly correlated with miR-146a-5p expression (R = 0.73, P = 2.29 × 10⁻⁴⁴), and with TMAO levels (R = 0.49, P = 1.84 × 10⁻¹⁵) (Fig. 6, D and E). Given the strong relationships between inflammation, miR-146a-5p, TMAO, and NF-κB, we propose a possible mechanism by which the three components may interact (Fig. 7). While we have yet to validate our proposed mechanism, the evidence of potential regulatory relationships between miR-146a-5p, TMAO, NF-κB, and Cd74 found in the literature and suggested by our results is noteworthy, and warrants further investigation.

Fig. 7.

Possible mechanism by which miR-146a is related to TMAO and inflammation.

DISCUSSION

High circulating TMAO levels are strongly associated with an increased risk for CVD. However, the genetic and molecular factors that are involved in regulating TMAO levels remain incompletely characterized. In this work, we’ve carried out analyses to identify a potential miRNA regulator of TMAO in a cohort of 288 female DO mice, of which half were fed an atherogenic diet. We also interrogated miRNAs in two additional animal models of atherosclerosis. The three main novel findings are the identification of: 1) miR-146a-5p as the only miRNA that is significantly upregulated and strongly correlated with TMAO in three independent animal models; 2) a novel QTL on Chromosome 12 that is associated with circulating TMAO levels; and 3) a link between miR-146a-5p and TMAO levels, possibly through upstream NF-κB activity and/or downstream miR-146a-5p-mediated regulation of Dlst and/or Numb.

Dlst (dihydrolipoyl succinyltransferase enzyme), a component in one of the rate-limiting enzyme complexes in the tricarboxylic acid cycle, has not been previously implicated in atherosclerosis or modulating TMAO levels. However, its deficiency has been shown to cause an increase in amyloid formation (12). Elevated amyloid fibril formation has been linked to Alzheimer’s disease and atherosclerosis (11, 30, 31). In addition, the regulatory relationship between miR-146a-5p and DLST has been recently validated as it relates to cardiac hypertrophy and dysfunction (20). Overexpression of miR-146a-5p in cardiomyocytes caused a decrease in DLST mRNA and protein. Given this connection, further investigation of the actions of Dlst on controlling TMAO and the onset or progression of atherosclerosis is warranted.

Numb has also been linked to Alzheimer’s disease (32). Moreover, it has also been directly implicated in the control of circulating cholesterol levels through its role in regulating intestinal cholesterol absorption (27, 44). A recent study reported that genetic variants in Numb may be associated with coronary artery disease in humans (1). Furthermore, it is worth noting that TMAO feeding has been shown to cause a decrease in cholesterol absorption and in Niemann-Pick C1-like1 (Npc1L1) abundance, which interacts with Numb during Clathrin-mediated cholesterol absorption in the intestine (23, 27). What, if any, functions Numb has in the liver, or in immune cells recruited to the liver in atherosclerosis, remains to be determined. Our study indicates that there are strong connections between Numb, TMAO, and atherosclerosis that are worth investigating further.

To our knowledge this study represents the first report of the potential relevance of the miR-146a-5p/Numb and miR-146a-5p/Dlst regulatory relationships in the context of cardiometabolic disease. Interestingly, miR-146a and miR-146b are members of the co-regulated module of miRNAs identified in our previous study that is associated with circulating LDL-C levels in the DO mice fed the HFCA atherogenic diet (10). In addition, Numb and Dlst are members of co-regulated modules that are inversely correlated with the LDL-C associated miRNA module (10). Our current and previous findings strengthen the burgeoning connection between miR-146 and diet-associated cardiometabolic disease by demonstrating a novel correlation with circulating TMAO via possible mediators of this relationship, Numb and Dlst. It is worth noting that the miR-146 target site in NUMB is conserved in humans; however, that of DLST is not (2, 18). It is also important to point out here that there are 80 genes in the Chr 12 TMAO QTL, and genes other than Dlst and Numb at this locus may be relevant to TMAO levels and atherogenesis and, therefore, merit attention in future studies as well.

The levels of miR-146a-5p are significantly elevated in atherosclerotic plaques and rise with disease progression (7, 34). Moreover, levels of miR-146 in serum have been associated with hyperlipidemia (37) and were shown recently to be reduced upon statin treatment (45). A very recent study using a knockout mouse model demonstrated that miR-146a-5p is a critical regulator of cholesterol metabolism and inflammatory signaling in the context of an atherogenic diet (7). The authors showed that genetic ablation of miR-146a-5p in bone marrow-derived cells reduces circulating LDL-C and suppresses atherogenesis, whereas deletion of miR-146a-5p in the vascular endothelium promotes the development of atherosclerosis, which is partly due to unrestrained inflammatory signaling. This complicated result points to the pleiotropism of miRNAs and their cell-type specific functionalities.

NF-κB is a prominent proinflammatory driver and a known transcriptional regulator miR-146a-5p expression (40). NF-κB is consequently negatively regulated by miR-146a-5p as part of a negative feedback loop. In addition, elevated TMAO levels have been shown to promote inflammation (28), and injection of TMAO into mice led to increased activation of inflammatory signals via NF-κB (35). In the DO mice, we have previously shown that there is an increase in circulating liver inflammation markers, AST (aspartate transaminase) and ALT (alanine transaminase) levels (10). Given these connections, we suggest that the positive correlation between miR-146a-5p and TMAO may be due to a compensatory response of miR-146a-5p to TMAO. Specifically, as TMAO levels rise and promote NF-κB-activated inflammation, miR-146a-5p expression may also increase in an attempt to mitigate it. Thus, miR-146a-5p may not be a driver of TMAO levels but, rather, a marker and an attempted response to quell the inflammation. There is still the question of the roles of Dlst and Numb, which are located in the TMAO QTL. These may be key players in the compensatory actions of miR-146a-5p against inflammation. It has been previously established that Numb is a regulator of proinflammatory cytokines with a connection to increased NF-κB activity (25). However, exactly how Numb and Dlst are contributing to the phenotype remains to be seen, and more investigation is needed to further elucidate this mechanism.

Overall, these findings strengthen the connection between miRNAs and dyslipidemia/atherosclerosis. We recognize that the conclusions drawn from our findings, however strong and well supported by studies in two other animal models, are associative; thus we propose further investigation to tease apart the regulatory mechanisms at work. Whether the rise in the levels of miR-146a-5p and other miRNAs contributes to atherogenesis, or is part of an adaptive response to atherogenic or inflammatory stimuli, is not yet known and merits deeper study.

GRANTS

This research was supported in part by National Institutes of Health Grants R01HL-128572 (B. J. Bennett), P30DK-056350 (B. J. Bennett and D. Pomp), R01DK-105965 (P. Sethupathy), a pilot grant from the Nutrition Research Institute (B. J. Bennett and P. Sethupathy), and a National Science Foundation Graduate Research Fellowship Program DGE-1144081 (A. R. Coffey). This work was supported by United States Department of Agriculture (USDA)/Agriculture Research Service/Western Human Nutrition Research Center project funds (2032-51000-004-00D). USDA is an equal opportunity provider and employer.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.R.C., M.K., T.L.S., J.A., W.P., R.Z.G., S.B.B., R.E.T., and B.J.B. analyzed data; A.R.C., S.B.B., D.P., P.S., and B.J.B. interpreted results of experiments; A.R.C. and M.K. prepared figures; A.R.C. drafted manuscript; A.R.C., M.K., P.S., and B.J.B. edited and revised manuscript; A.R.C., P.S., and B.J.B. approved final version of manuscript; T.L.S., J.A., K.H., E.G., S.B.B., R.E.T., and B.J.B. performed experiments; P.S. and B.J.B. conceived and designed research.