The distribution of circulating microRNA and their relation to coronary disease

Background: MicroRNAs (miRNAs) are small RNAs that regulate gene expression by suppressing protein translation and may influence RNA expression. MicroRNAs are detected in extracellular locations such as plasma; however, the extent of miRNA expression in plasma its relation to cardiovascular disease is not clear and many clinical studies have utilized array-based platforms with poor reproducibility. Methods and Results: Initially, to define distribution of miRNA in human blood; whole blood, platelets, mononuclear cells, plasma, and serum from 5 normal individuals were screened for 852 miRNAs using high-throughput micro-fluidic quantitative RT-PCR (qRT-PCR). In total; 609, 448, 658, 147, and 178 miRNAs were found to be expressed in moderate to high levels in whole blood, platelets, mononuclear cells, plasma, and serum, respectively, with some miRNAs uniquely expressed. To determine the cardiovascular relevance of blood miRNA expression, plasma miRNA (n=852) levels were measured in 83 patients presenting for cardiac catheterization. Eight plasma miRNAs were found to have over 2-fold increased expression in patients with significant coronary disease (≥70% stenosis) as compared to those with minimal coronary disease (less than 70% stenosis) or normal coronary arteries. Expression of miR-494, miR-490-3p, and miR-769-3p were found to have significantly different levels of expression. Using a multivariable regression model including cardiovascular risk factors and medications, hsa-miR-769-3p was found to be significantly correlated with the presence of significant coronary atherosclerosis. Conclusions: This study utilized a superior high-throughput qRT-PCR based method and found that miRNAs are found to be widely expressed in human blood with differences expressed between cellular and extracellular fractions. Importantly, specific miRNAs from circulating plasma are associated with the presence of significant coronary disease.

been reported due to the limitations of current arrays and the cost of extensive qRT-PCR screening. Using blood samples from normal subjects and high-throughput qRT-PCR, the miRNA expression profile was determined for whole blood, isolated platelets, mononuclear cells, plasma and serum from five healthy subjects. Of the 852 miRNAs measured (Supplemental Table 1), the distribution of circulating miRNA that were most abundantly expressed in plasma and many of the blood derived sources is shown in Table 1. Gene expression is listed as cycle threshold value (Ct) consistent with RT-PCR-based data. Because the Ct values are listed, higher gene expression is reflected by a lower number. A complete list of miRNA expression for all sources is shown in Supplemental Table 2 (mean Ct value) and Supplemental Table 3 (mean delta Ct, accounting for housekeeping gene hsa-RNU1A-1). Unlike mRNA expression, it is currently unclear if miRNA data is more reliable when normalized with a housekeeping gene. This is especially germane for the cell-free plasma samples where a fixed volume was used and gene expression does not need to be normalized for cell count.

Introduction
MicroRNAs (miRNAs) are 20-26-nt single stranded RNAs that participate in the regulation of various biological functions in numerouseukaryotic lineages, including plants, insects, vertebrate, and mammals [1][2][3] . More than 850 human miRNAs have been cloned and bioinformatic predictions indicate that mammalian miRNAs can regulate approximately 30% of all protein-coding genes [4][5][6] . The expression of many miRNAs is specific to a tissue or developmental stage, and the miRNA expression pattern is altered during the development of several diseases 7,8 .
Studies measuring small numbers of miRNAs have shown their presence in circulating blood; specifically in platelets 9 , plasma 10 , and mononuclear cells 11 . In studies examining specific miRNAs 12,13 , differential expression was noted both in hematopoietic cell lines 13 and between human and mouse cells 12 . Interestingly, miRNAs have been detected in cell-free human plasma preparations 14 . They have been found to be stable and protected from endogenous RNase activity 14 . In addition, levels of a specific miRNA (miR-141) can distinguish patients with prostate cancer from healthy controls 14 .
Basic studies have shown that miRNAs regulate cardiac differentiation, angiogenesis, and myocyte growth 15,16 . Small clinical studies have also shown that levels of specific miRNA have been correlated with myocardial infarction in cardiac tissue from humans 17 and animal models 18,19 . A recent study examined one miRNA (miR-1) from plasma and related it to acute myocardial infarction 10 . In stable and unstable coronary artery disease patients, 157 miRNAs were measured from peripheral blood mononuclear cells and differential expression was found 11 .
As shown by these studies, there are some publications about circulating miRNAs in cardiovascular disease 9,20 . In addition, the existing studies are restricted by incomplete evaluation of currently known miRNAs. Available arrays are constrained by incomplete miRNA coverage, issues in discriminating between closely related miRNAs, as well as ongoing discovery of new miRNAs (and the inherent lack of flexibility of the array platform). In addition, miRNA arrays have poor reproducibility when across clinical populations or in larger samples sizes 21 . We have developed a combined method of miScript miRNA assays (Qiagen, Germantown, MD) and Dynamic Arrays (Fluidigm, South San Francisco, CA) that employs quantitative RT-PCR (qRT-PCR) using a high-throughput process that allows us to analyze more samples vs. more miRNAs in a very short time 22 . Lastly, unlike hybridization-based microarray profiling techniques, qRT-PCR is considered the gold standard for RNA expression and does not require further confirmation analysis. Using this platform, a complete analysis of circulating whole blood, cellular, and cell-free miRNA was performed. The relevance of these findings to coronary disease was determined by measuring plasma miRNA expression in patients presenting for coronary angiography. The findings suggest that the distinct patterns of miRNA expression in components of whole blood may reflect specific patterns of disease.

Results
miRNA profile in human plasma, serum, platelets, mononuclear cells and whole blood There is minimal information defining miRNA expression in human blood and a complete screen of all known miRNAs has never miRNA results and demographic data One-hundred and ninety four miRNAs were not expressed in any of the blood-based samples. Peripheral blood mononuclear cells (PB-MCs) contained the highest number of miRNAs (658) followed by whole blood (609), platelets (448), serum (178) and plasma (147). As the abbreviated list shown in Table 1 demonstrates, while there between plasma and cellular miRNA patterns. Based on the significant overlap between the groups, it is difficult to determine the source for many of the specific plasma-derived miRNAs. It is not clear from the current data which miRNAs have a non-blood-based cellular source. Interestingly, miR-1185 and miR-548a-5p were much more abundantly expressed in plasma as compared to PBMCs or platelets, or only expressed in plasma. Although an interesting observation, the source of these two miRNAs cannot be determined from this data. In addition, some genes were expressed in platelets or PBMCs and not whole blood. This is likely due to the greater dilution used with PAXgene tubes and the loss of measurable expression of less abundant miRNAs.

Plasma miRNA expression in patients with and without significant coronary disease
The initial hypothesis of this study was that patients with significant coronary disease would have altered expression of plasma miRNA. Patients were divided into two groups ( Table 2); 1) ≥70% coronary stenosis of any coronary artery or 2) <70 coronary stenosis. There were notable differences in expression of some of the miRNAs between these two groups (Table 3). By direct comparison, several plasma miRNAs were found to have over 2-fold increased expression in patients with significant coronary disease (≥70%) as compared to those with minimal coronary disease or normal coronary arteries ( Figure 1). Initial statistical analysis demonstrated that antihypertensive therapy, smoking, and lipid lowering therapy have a positive association with coronary artery status ( Table 2). Increased expression of miR-494, miR-769-3p and miR-490-3p was associated with ≥70% coronary stenosis. Next, the six variables were fit into a logistic regression analysis. As seen in Table 4, anti-hypertensive therapy, smoking and miR-769-3p were significantly associated with the coronary status. is consistency between the groups, some miRNAs are more highly expressed in select sources. Sixty miRNAs were expressed in all five components. As the precise source of miRNA in plasma is not yet known, there was particular interest in comparing the expression   Fold change in plasma miRNA expression between three groups. miRNAs shown had expression increased >2-fold or decreased <0.5-fold for the groups; 1) patients with CAD; at least one of the coronary arteries have ≥70% occlusion; 2) patients with atherosclerosis but not clinically significant CAD; coronary occlusion 1%-<70%; and 3) coronary atherosclerosis-free, patients with no angiographically documented coronary artery stenosis.  We initially determined whether presence of significant coronary disease, as defined by ≥70% stenosis is associated with specific miRNA expression. A secondary question is whether presence/ absence of coronary disease is associated with miRNA expression. To study this, patients were placed into one of three groups; 1) patients with CAD; at least one of the coronary arteries have ≥70% occlusion; 2) patients with minimal CAD; coronary occlusion 1%-<70%; and 3) coronary atherosclerosis-free, patients with no angiographically documented coronary artery stenosis. Because the numbers are small, a simple (non-statistical) comparison was made to determine trends. As seen in Figure 2 and Supplemental table 4, 18 miRNAs had a varied expression pattern between group 1 and group 3. Seventeen miRNAs were upregulated at least 2-fold with only miR-1914 downregulated 0.5-fold. Interestingly, most miRNAs demonstrated a dose response with the greatest expression in patients with the most coronary disease and the least expression in disease-free patients (Supplemental table 4).

miRNA target predictions
Currently, the capacity to measure miRNAs far outpaces our ability to understand their function in a given tissue. However, to better understand the potential significance of our findings, we conducted analyses that predict targets of the miRNAs that were up-or downregulated in coronary disease using two methods; current publications (www.ncbi.nlm.nih.gov/pubmed/) and TARGETSCAN 5.1 (www.targetscan.org). Using these methods, results for miRNA targets varied between 3 to 369 targets per miRNA when identified. With the TARGETSCAN search, target genes for individual miRNAs were identified using the context score for specific sites within genes. The context score is the sum of site-type contribution, 3' pairing contribution, local AU contribution, and position contribution. The lower the context score indicates the most highly predicted targets for each miRNA. By TARGETSCAN search, miR-NAs miR-1914 and miR-7-2, had no target gene identified.
Detailed predictions for the miRNAs found to be significant in CAD are shown in Table 5. In broad terms, some miRNAs appear to target transcription factors, growth factors, cytokine regulation, transmembrane proteins, signal transduction pathways, and epigenetic pathways such as histone acetylases. Prediction results for specific miRNAs in TARGETSCAN include the following: miR-129-3p and miR-494 target HMGCS1; miR-150 targets MMP14; miR-150, and 92b target MMP16. Additionally, miR-1207 appears to target energy metabolism (most likely involving glucose metabolism).
We also analyzed the miRNAs that were unique to plasma, miR-1185-1 and miR-548a-5p-1. While many potential targets were identified, these miRNAs were predicted to target genes involved in controlling transcription factors, as well as several growth and cell cycle components.

Discussion
MicroRNAs (miRNAs) are short regulatory RNAs that participate in the control of approximately 30% of all protein-coding genes [4][5][6] . The expression of many miRNAs is usually specific to a tissue or developmental stage, and the miRNA expression pattern is altered during the progression of several diseases 7,8 . Most miRNAs are transcribed by RNA polymerase II from individual miRNA genes, from introns of protein coding genes, or from polycistronic transcripts that often encode multiple related miRNAs 4,23 . Although Table 5. MiRNA predictions for target genes associated with significant coronary disease using TARGETSCAN. Individual miRNAs target genes identified as significantly different between patients with and without significant coronary disease were selecting using the context score, a sum of; site-type contribution, 3' pairing contribution, local AU contribution, and position contribution. The lowest scores show highest predictions. leukocyte and platelet recruitment to the atherosclerotic tissues, matrix reorganization, foam cell formation, growth/proliferation, and angiogenesis. However, these predictions do not provide definitive targets and additional basic studies are needed to provide clearer mechanistic information.
Also unique to our study is the specific method of measurement we used, which allowed for flexibility in adding newly discovered miR-NAs, the use of small volumes, and high-throughput methods for qRT-PCR. Currently, there are several miRNA microarray products available that measure fewer miRNAs and some consist of older versions of the Sanger miRBase Sequence Database. Using the universal cDNA reaction feature of miScript provided the ability to profile all miRNAs with one cDNA reaction. Unlike the hybridization-based microarray profiling techniques, by coupling the miScript and Biomark Systems, confirmation analysis was not required for individual miRNAs. However, there are important limitations of our study. Despite the expansive miRNA survey for blood components, we cannot define the specific source for the plasma miRNA nor its eventual destination. Our study of coronary patients was limited by only being able to evaluate plasma miRNA, as other blood components were not available to us. In addition, while the miRNA data provided from these patients are unique, the numbers are still small making further analysis based on any subgroup statistically unfeasible.
In summary, miRNAs are small RNAs that play an important role in the negative regulation of gene expression by suppressing protein translation and have been detected in cell-free plasma and have been related to select diseases. By examining all measurable miR-NAs, we defined the relative expression in blood components and find significant expression in platelets, PBMCs, whole blood, plasma and serum. By comparing plasma miRNA expression in patients with coronary disease, we begin to define specific miRNAs that are altered and provide potential targets that influence atherosclerosis.

Sample procurement
This study was approved by Mersin University Ethical Committee (06/05/2009, #6/144) and written consent was obtained from the subjects to test the hypothesis of whether coronary occlusion of ≥70% is associated with increased plasma miRNA expression levels. Upon enrollment, a study coordinator identified the presence of the following risk factors: (1) age, (2) male sex, (3) clinical history of diabetes, (4) clinical history of hypertension, (5) cigarette smoking, clinical history of hypercholesterolemias, and (7) family history of coronary disease. Coronary angiograms were analyzed off-line in a blinded fashion with the use of digital calipers to measure stenosis severity, and stenosis was defined as a dichotomous variable: if a stenotic lesion was ≥70%, that vessel was counted as stenosed.
The presence or absence of stenotic disease was also noted. Patients were ranked as having 0-to 3-vessel disease (number of coronary arteries with detectable atherosclerotic disease). For each patient, K 3 EDTA arterial blood (5 mL) was collected just prior to coronary angiography. Blood samples were centrifuged (3,000 g) and 400 µl plasma samples were stored at -80°C until RNA isolation.
In a separate smaller study, blood was obtained from healthy consented volunteers (n=5; 3 female, 2 male, average age=45) at miRNAs can guide mRNA cleavage, the basic function of miRNA is to mediate inhibition of protein translation 1,7,[24][25][26][27] through miR-NA-induced silencing complexes (miRISCs). The guiding strand of miRNA in a miRISC interacts with a complementary sequence in the 3'-untranslated region (3'-UTR) of its target mRNA by partial sequence complementarities, resulting in translational inhibition 1,7 .
In this study, the distribution of miRNA expression in whole blood, platelets, PBMCs plasma and serum showed significant overlap. Of particular interest are the nucleus-lacking platelet and the cell-free plasma expression levels. A primary question is why platelets would have miRNA? Platelets are produced in the bone marrow from megakaryocytes as cytoplasmic fragments without genomic DNA 28 . Platelets, however, retain a small amount of megakaryocyte-derived messenger RNAs (mRNAs) that have recently been suggested to be of physiological significance. Platelets can respond to physiological stimuli at the levels of protein translation and mRNA splicing 29,30 . There are few published studies describing platelet miRNAs 13 . In this study, cells of hematopoietic lineage were described to have a limited number of miRNAs (this study only tested for 19 miRNAs) and functionality was not shown. Interestingly, our data demonstrate that platelets express nearly the same number of miRNAs as PBMCs. The number of miRNAs in PBMCs is slightly less than in whole blood with the reason likely being dilution of low abundant PBMC miRNAs in whole blood.
In limited numbers, miRNAs have been detected in cell-free human plasma preparations 14 . They have been found to be stable and protected from endogenous RNase activity 14 . In addition, levels of a specific miRNA (miR-141) can distinguish patients with prostate cancer from healthy controls. In our analysis, we found moderate to high levels of expression of miRNAs in plasma in both normal subjects and patients with coronary disease, albeit in lower numbers as compared to platelets and PBMCs. What these data do not provide is the specific source of circulating miRNAs. This is a fundamental and fascinating question. They are believed to arise from three potential mechanisms: apoptosis, cellular activation with release of protrusions, and microsome/microvesicle formation. For example, Mitchell et al. found miR-141 differentially expressed in plasma in microsomes of prostate cancer patients 14 . It is possible that the miR-NAs detected by our measurements in plasma or blood could be derived from endothelial cells or the atherosclerotic plaque itself.
Further fundamental experiments are needed to answer this question. Recently, it has been shown that HDL particles deliver miR-NAs 31 . Additionally, Wang et al. reported that plasma and whole blood miR-133 and miR-328 levels are increased in AMI patients 32 .
By the current data, we cannot assign precise punitive targets; however, specific bioinformatics approaches have been developed to predict miRNAs present in the genome of different organisms. These techniques are based on the observation that transcripts are usually highly conserved between related species and produced from precursor transcripts of similar size and structure. Using these bioinformatics approaches and the limited information available in the literature, we assembled potential target genes for the miRNAs expressed in significantly different amounts between patients with and without ≥70% coronary disease ( Table 5). The list included a diverse range of functional and structural genes. This includes Boston University School of Medicine as previously described 33 . The study was approved by Boston University IRB and written consent obtained from the subjects. All subjects were free of medications or supplements, and had no history of hypertension, diabetes, smoking, or hyperlipidemia. Blood was collected into PAXgene RNA tubes (Becton Dickinson, Franklin Lakes, NJ) for whole blood, into CPT tubes (Becton Dickinson) for peripheral blood mononuclear cells (PBMCs), into citrate tubes for platelets and plasma, and empty tubes for serum. Isolated platelets were prepared as previously described.

Statistics
When examining miRNAs that had larger fold changes between the CAD groups ( Figure 1 and Figure 2) and those more highly expressed in the circulation using similar bioinformatics methods (such as targetscan), there were many miRNA targets that are known to control the processes important in the development of atherosclerosis (list not shown).

Univariate and bivariate analysis
We initially summarized our data in different stratifications based on our outcome variables (coronary disease status). Next, we examined the bivariate relationship between the response variable and quantitative covariates using either two-sample t-test or Kruskal-Wallis test, where appropriate. Specifically, the t-test was conducted to test for a mean difference in quantitative demographic variables and miRNA expression level between two categories of coronary disease status. The pooled variance or the Satterthwaite's method was used to estimate variance based on the equality of variance test. We employed Kruskal-Wallis test to determine whether there was any mean difference among groups in the scenario with three-category outcome variable.

Multivariate analysis
Tree-based methods have been increasingly applied to biological research such as microarray data analysis and genome-wide association studies 34,35 . RandomForest is a flexible nonparametric approach, which consists of many decision trees from bootstrap samples. In this study, we constructed randomForest 25 in identifying relevant variables to our outcome variable using the random-Forest package in R (2.10.1) 36 . Furthermore, we also conducted logistic regression and multinomial logistic regression where appropriate 37 .