Regulation of Coagulation Factor XI Expression by MicroRNAs in the Human Liver

High levels of factor XI (FXI) increase the risk of thromboembolic disease. However, the genetic and environmental factors regulating FXI expression are still largely unknown. The aim of our study was to evaluate the regulation of FXI by microRNAs (miRNAs) in the human liver. In silico prediction yielded four miRNA candidates that might regulate FXI expression. HepG2 cells were transfected with miR-181a-5p, miR-23a-3p, miR-16-5p and miR-195-5p. We used mir-494, which was not predicted to bind to F11, as a negative control. Only miR-181a-5p caused a significant decrease both in FXI protein and F11 mRNA levels. In addition, transfection with a miR-181a-5p inhibitor in PLC/PRF/5 hepatic cells increased both the levels of F11 mRNA and extracellular FXI. Luciferase assays in human colon cancer cells deficient for Dicer (HCT-DK) demonstrated a direct interaction between miR-181a-5p and 3′untranslated region of F11. Additionally, F11 mRNA levels were inversely and significantly correlated with miR-181a-5p levels in 114 healthy livers, but not with miR-494. This study demonstrates that FXI expression is directly regulated by a specific miRNA, miR-181a-5p, in the human liver. Future studies are necessary to further investigate the potential consequences of miRNA dysregulation in pathologies involving FXI.


Introduction
Although coagulation factor XI (FXI) was discovered nearly 50 years ago [1], its role in pathophysiological conditions is still not fully understood. A wide range of FXI plasma levels has been found in the healthy population [2]. The available functional data on FXI function are confusing, probably reflecting the fact that FXI might be involved not only in haemostasis but also in pathologic processes as inflammation or innate immunity [3,4]. Epidemiological and animal model studies have associated FXI levels with the risk of thrombotic disease (for review see [5,6]) or septic survival advantage [7]. On the other hand, FXI deficiency does not usually lead to spontaneous bleeding, but it is associated with an increased risk of bleeding when the haemostatic system is challenged [6,8]. Moreover, FXI inhibition has been proposed as a novel approach to developing new anti-thrombotic therapies to achieve an improved benefit-risk ratio [9,10].
In this framework, several groups have been engaged in an intensive study of the influence of genetic and environmental factors on FXI plasma levels in an attempt to understand whether the heterogeneous values found in the healthy population confer a pro-or anti-thrombotic phenotype. Although some of these studies have identified the involvement of common single nucleotide polymorphisms in the structural F11 gene and alterations in other genes that might indirectly regulate plasma levels of this factor [11][12][13], the molecular mechanisms of FXI regulation are still largely unknown.
MicroRNAs (miRNAs), which are small non-coding RNAs that regulate protein expression [14], have been involved in the regulation of many complex mechanisms or physiological conditions, including the haemostatic system. Available predictive algorithms estimate that a third of the human mRNAs may contain a single or multiple binding sites for miRNAs [15]. As such, a single miRNA can potentially target hundreds of genes, or a single gene could be targeted by many different miRNAs [15][16][17][18]. However, it has been shown that overexpression of miRNAs only provokes a mild repression of both mRNA [19] and protein [20].
During the last four years, several groups including ours, have evaluated the role of miRNAs in the regulation of haemostasis [21]. Coagulation factors like fibrinogen [22] or tissue factor have been described as interacting with miRNAs, which may have an impact on the thrombotic etiology associated with pathologies such as antiphospholipid syndrome or systemic lupus erythema-tosus [23]. Recently, PAI-1 [24] and protein S [25] have also been shown to be directly regulated by miRNAs. In the current study, we investigated the potential relevance of miRNAs aiming to discover new elements that may modulate FXI in the liver. In silico predictions together with in vitro experiments showed that only miR-181a-5p caused a slight although significant decrease both in FXI protein and F11 mRNA levels. Luciferase assays helped us to demonstrate a direct interaction between miR-181a-5p and 39 untranslated region (39UTR) of F11. Importantly, F11 mRNA levels were inversely and significantly correlated with miR-181a-5p levels in 114 healthy livers. This study demonstrates that FXI expression in the human liver is directly regulated by a specific miRNA, miR-181a-5p, opening up new prospects in a better understanding of the pathophysiology of haemostatic diseases where FXI is involved and in the development of miRNA-based therapeutic technologies.

Results
A microarray and in silico target search yielded four miRNAs that could potentially bind to F11 mRNA In order to select miRNAs with the potential to bind to F11 mRNA, two criteria were established (i) the miRNA expression cut-off in liver had to be .500 arbitrary units (au) (see array in Table S1) and (ii) the miRNA binding had to be anticipated in 4 or more of the prediction algorithms of miRNA targets used (n = 8). Such filtering allowed the selection of four miRNAs: miR-181a-5p (liver expression = 1233 au; 6 prediction algorithms), miR-23a-3p (liver expression = 6052 au; 5 prediction algorithms), miR-16-5p (liver expression = 3513 au; 4 prediction algorithms), and miR-195-5p (liver expression = 3046 au; 4 prediction algorithms) (see Table 1 and Table S1). Additionally, a negative control miRNA (miR-494; see Table 1), which was not predicted to bind to F11 mRNA (only 2 prediction algorithms) and with a liver expression .500 au, was also investigated (Table 1). Putative binding sites for miRNAs are shown in Figure 1. Whereas miR-181a-5p and miR-23a-3p bind to two closely located sites, miR-16-5p and miR-195-5p share the same binding site located ,200 bp downstream of the miR-23a-3p seed ( Figure 1).
In vitro studies suggested miR-181a-5p as a direct inhibitor of FXI and F11 mRNA expression To test which miRNAs may inhibit FXI expression, we employed HepG2 cells, expressing lower levels of these miRNAs than the liver ( Figure S1A). Transfection of HepG2 cells with the different miRNA mimics showed that only miR-181a-5p mimic provoked a significant reduction of endogenous F11 mRNA levels of almost 30% (100% vs. 7169%; p = 0.03; N = 3) compared with non-specific scrambled negative control (SCR) transfection ( Figure 2A). No inhibition was found when transfecting HepG2 with the other selected miRNAs or with miR-494 ( Figure 2A). In fact, miR-181a-5p caused a significant decrease (,30%) in the levels of extracellular FXI (100% vs. 7167%; p = 0.04; N = 3) ( Figure 2B).
Next, we tested the effect of a miR-181a-5p inhibitor in another hepatic cell line where miR-181a-5p expression levels were similar to those reported in the liver ( Figure S1B), PLC/PRF/5 hepatic cell line. Our results indicated that inhibition of miR-181a-5p increased both the levels of F11 mRNA (100% vs. 12163%; p = 0.006; N = 3) ( Figure 2D) and of extracellular FXI (100% vs.  The presence of the selected miRNA in the target prediction tool is recognized in the 11666%; p = 0.038; N = 3) ( Figure 2E) further supporting a physiological regulation of FXI by miR-181a-5p. Whether the lower levels of F11 mRNA observed in transfected HepG2 were due to an indirect effect of miR-181a-5p or to mRNA decay was further investigated. Co-transfection of HCT-DK cells with a luciferase reporter vector containing the 39UTR of F11 and miR-181a-5p showed a significant decrease of ,30% of the luciferase activity in comparison with SCR (100617% vs. 71627%; p = 0.04; N = 3). This inhibition was not observed when using a mutated vector which lacked the binding site of miR-181a-5p ( Figure 3).

F11 mRNA and FXI levels were inversely correlated with miR-181a-5p in human livers
In order to establish the potential physiological significance of the above results, we measured F11 mRNA and miR-181a-5p levels in samples from a cohort of healthy livers that had been used for liver transplant. Using a linear regression model, F11 mRNA levels were found to be inversely and significantly related to miR-181a-5p levels (r = 20.184; p,0.05) ( Figure 4A). To confirm the specificity of this result, we compared levels of F11 mRNA with those of miR-494 (that showed no in silico or in vitro effect on FXI expression) and we found no association between these two molecules (r = 20.033; p = 0.727) ( Figure 4B).
Previous studies have shown that FXI plasma antigenic levels are correlated with those of FIX [2]. In this study, we observed a strong correlation between F9 and F11 mRNA levels ( Figure 4C), which supported the described correlation in plasma. Aiming to further confirm the specificity of F11 mRNA: miR-181a-5p interaction, we investigated a potential correlation between miR-181a-5p and F9 mRNA levels. As expected, our results demonstrated that miR-181a-5p had no influence on F9 mRNA levels in healthy livers (p = 0.115) ( Figure 4D).
Next, we investigated the dynamic range of expression of F11 mRNA in hepatocytes, finding that it was larger than that described for plasma levels of FXI [26]. More specifically, there was a three-fold difference in F11 mRNA values between the 75 th and 25 th percentiles ( Figure 5A). Interestingly, we also observed a wide range of expression of miR-181a-5p among individuals. In fact, miR-181a-5p expression in liver samples with F11 mRNA levels below 15 th percentile was two-fold higher than in the samples above the 85 th percentile ( Figure 5B).

Discussion
It has been consistently reported that miRNAs may regulate haemostatic proteins such as fibrinogen [22], tissue factor [23], PAI-1 [24] or antithrombin [21], while variations in the levels of miRNAs [23] or in the efficacy of the miRNA: mRNA interaction [27] might have an impact on the development of thrombotic diseases. In the present study, we provide new evidence that miRNAs fine-regulates the expression of FXI in human hepatocytes, which in turn opens an alternative regulation pathway that may be exploited for future studies.
The in silico results were the starting point in our work. Bearing in mind that approximately 70% of all in silico predictions are thought to be false-positive [28], this encouraged us to more accurately establish filters to select the candidate miRNAs to be studied. Our filters led to the selection of 4 candidate miRNAs with potential to bind to F11 mRNA and, even then, we observed 75% false positives. The effective prediction of miRNA: mRNA pairs in animal systems is still a challenge because of the complexity of this interaction [29]. Future studies to unravel these issues will probably help identify a more selective in silico search method for additional miRNAs which target FXI and other haemostatic factors.
In a second step consisting in performing an in vitro validation, only miR-181a-5p was potentially a direct regulator of F11 expression, which enabled it to induce a significant decrease in both extra-and intracellular FXI and F11 mRNA levels. As expected, and in accord with other studies describing that the effect of a given miRNA on its target is generally modest [19,20,28] and that it is the total of various miRNA interactions that would determine the effect as a whole, we found that the effect of miR-181a-5p on FXI expression was mild. Future studies will determinate additional miRNAs that may act in conjunction with miR-181a-5p to regulate FXI expression. In this sense, we found that a neutral miRNA (miR-494) neither interacted nor correlated with F11 mRNA ( Figure 4B). Moreover, the lack of correlation found between miR-181a-5p and F9 mRNA further supported our hypothesis that miR-181a-5p has a specific effect on FXI expression.
Next, we investigated the physiological relevance of FXI regulation by miR-181a-5p. Given our inability to test this hypothesis in vivo, we performed ex vivo analysis in livers from healthy donors that had been used for transplant. Our data showed that levels of miR-181a-5p and F11 mRNA were correlated in human livers (Pearson's coefficient = 20.184, p,  inhibitor (anti-181a) or SCR inhibitor (Inh-SCR). Results are represented as mean 6 SD of three replicates from three independent experiments. The normalized data were expressed as changes relative to the data of the cells transfected with SCR mimic or SCR inhibitor and set as 100%. *P,0.05. Student's t-test was calculated in mimic precursors vs. SCR. doi:10.1371/journal.pone.0111713.g002 0.05). In fact, the in vitro study demonstrated that the decrease of both target F11 mRNA and protein levels was proportional and therefore we speculate that the ex vivo correlation may be extrapolated. Together with the fact that protein levels are determined by additional tightly modulating processes including protein degradation rates, the final effect of a miRNA on an mRNA is very difficult to predict [30]. In this sense, we found a three-fold difference in F11 mRNA values between the 75 th and 25 th percentiles in healthy livers, whereas only a 1.3-fold difference in plasma between the same percentiles has previously been found [26]. Therefore, translational, post-translational, and degradation processes, together with the regulation of the secretory pathway (involving both miRNAs and target genes), may act to adjust the final amount of FXI in plasma.
Many observations of miRNA-mediated regulation in mammalian cells considered them as fine-tuners of gene expression [31]. Indeed, miRNA expression can be regulated by the same genetic alterations that modulate protein coding genes [32] as well as by environmental factors [6]. Specifically, miR-181a-5p has been shown to be regulated by dopamine [33] and TGF-b [34]. Our data showed a surprising heterogeneity of miR181a-5p expression in human liver samples ( Figure 5A). We observed a 2.5-fold difference in miRNA values between the 75 th and 25 th percentiles, which reached more than 12-fold when the 90 th and 10 th percentiles were compared. Interestingly, recently Mendell and Olson suggested that miRNA deficiency or overexpression may have an important impact under pathophysiological stress conditions [35] while, in physiological conditions, miRNAs may only play a modest role in regulating their target.
Overall, our in vitro and ex vivo results establish miRNAs as new modulators of FXI, opening up new prospects for the regulation of FXI by miRNAs that deserves further attention and confirmation. However, our in vitro experimental conditions did not allow us to test the effect of miR-181a-5p on functional activity of FXI. Additional studies are necessary to fully understand this new FXI regulation and to test the possibility that it is regulated by other miRNAs in an indirect way. It would also be useful to further investigate the association of miR-181a-5p expression with the development of thromboembolic disease. In this sense, FXI is seen as a potential therapeutic target since its inhibition prevents thrombosis without bleeding episodes [10]. Indeed, the use of several miRNA-based therapies in the liver is under investigation. The most advanced is the use of anti-miR-122 in the treatment of hepatitis C [36]. On the other hand, it has already been shown that antisense oligonucleotides inhibit factor XI expression in vivo [37], and a clinical phase 2 trial is currently ongoing (http://www. isispharm.com/Pipeline/Therapeutic-Areas/Cardiovascular.htm #ISIS-FXIRx) using a drug antisense in patients with total knee arthroplasty. Therefore, the characterization of miR-181a-5p that regulates FXI expression may be seen as an opportunity to start envisaging the potential use of miRNA precursors as an antithrombotic drug.  (Table S2). All donors gave a written informed consent that was recorded following the procedures of each Biobanc. Human liver studies were further approved by Local Ethics Committee from Hospital Universitario Morales Meseguer in Murcia (#ESTU-19/12).

MiRNA array and in silico identification of miRNA binding sites in F11 mRNA
To identify miRNAs expressed in healthy liver, we performed expression arrays including 1,898 human mature miRNAs (LCSciences, Houston, TX; Sanger mirBase Release 18.0 and 19.0) using total RNA extracted from 4 healthy human liver samples (Table S1). Raw miRNA microarray data are available in public archives under corresponding author name (GSE61219). . Luciferase reporter assays. Schematic diagram of the luciferase reporter plasmids including F11 WT 39UTR or F11 mutant 39UTR in which the seven nucleotides forming the seed region of miR-181a-5p were deleted. HCT-DK cells, that do not express miR-181a-5p nor other Dicer-dependent miRNAs that may interfere in miRNA overexpression experiments, were transfected with either F11 WT 39UTR or F11 mutant 39UTR along with 100 nM miR-181a-5p precursor. A SCR precursor was used as control. Luciferase activities were normalized to renilla activities. Results are represented as mean 6 SD of three replicates from three independent experiments. The normalized data were expressed as changes relative to the data of the cells transfected with SCR mimic and set as 100%. *P,0.05. Student's t-test was calculated in mimic precursors vs. SCR. doi:10.1371/journal.pone.0111713.g003 For the determination of mature hepatic miRNAs potentially regulating human F11 39UTR, we used four miRNA prediction algorithms.

Cell transfection
To validate in silico experiments, we performed transfection assays in cell lines mentioned above. Briefly, cells were seeded twenty four hours before transfection in complete medium without antibiotics and transfected with 100 nM of chemically modified double-stranded RNAs that mimic endogenous miRNAs (mimic), 100 nM miRNA inhibitors (miRCURY LNA microRNA Inhibitor from Exiqon, Vedbaek, Denmark) or 100 nM SCR from Life Technologies (Madrid, Spain) as previously described [40]. Transfection efficiency was .90% ( Figure S2). After 48 h, supernatants and cells were collected for subsequent mRNA and protein analyses.

MiRNA and mRNA expression levels
Total RNA was isolated from both fresh livers and transfected cells using Trizol Reagent (Life Technologies, Madrid, Spain). RNA (400 ng) and SuperScript III First-Strand Synthesis System (Life Technologies, Madrid, Spain) were used for reverse transcription (RT) reactions. F11 and ACTB (as endogenous reference control) genes expression were quantified by qRT-PCR (Hs01030011_m1 and Hs99999903_m1, respectively, from Life Technologies, Madrid, Spain).

Western blot
Proteins from the lysate of transfected HepG2 cells (60 mg) or liver (50 mg) were blotted and immunostained with anti-human FXI polyclonal antibody (Enzyme Research Laboratories, Swansea, UK) and anti-human b-actin monoclonal antibody (Sigma-Aldrich, Madrid, Spain). Additionally, we collected and lyophilized 500 mL supernatants from HepG2 and PLC/PRF/5 using CentriVap Concentrator (Labconco, Kansas, MO) and 24 mL were blotted and immunostained with anti-human FXI polyclonal antibody. FXI and b-actin were immunodetected with the appropriate secondary antibody labeled with peroxidase (GE Healthcare, Barcelona, Spain). Detection was performed using ECL Prime Western Blotting Detection Kit (GE Healthcare, Barcelona, Spain) and ImageQuant LAS 4000 Imager. Densitometric analysis was performed with ImageJ software (http://rsb. info.nih.gov/ij/). Data were expressed as changes relative to the values of the cells transfected with SCR, taken as 100%.

Luciferase reporter assay
In order to test if the regulation of F11 by miRNAs is done through a direct interaction between both molecules, we performed luciferase reporter assays, as described below.
To generate mutations in the predicted target site for miR-181a-5p, seven nucleotides (TGAATGT) located in the seed sequence were deleted using the QuikChange site-directed mutagenesis kit (Agilent Technologies, Santa Clara, CA). In silico prediction, using RNAHybrid, of miR-181a-5p binding to the mutated sequence showed a 33% decrease in the minimum free energy value (not shown) indicating that miR-181a-5p: F11 mRNA interaction was completely suppressed. Sequencing was performed to check for the deletion of the seed sequence. The primers used (del_181_AS and Del_181_S) are detailed in Table  S3.
Luciferase vector transfection. HCT-DK cells, that do not express miR-181a-5p, were seeded at a density of 80,000 cells/well in 24-well plates with complete McCoy's 5A supplemented with 10% fetal calf serum without antibiotics. The following day, cells were co-transfected with miR-181a-5p (both pMIR-REPORT plasmids -1000 ng/well-wild type or mutated for the miR-181a-5p seed site) or SCR precursor, and 100 ng/well of renilla luciferase control plasmid (pRL-TK; Promega, Madison, WI) using Lipofectamine LTX (Life Technologies,Madrid, Spain) according to the manufacturer's instructions. Luciferase assays were performed as previously described [40].The enzymatic activities of renilla and firefly luciferases were quantified in a Synergy 2 luminometer (Biotek, Winooski, VT). Each combination of pMIR-REPORT (wild-type and mutated 39UTR) and pRL-TK was tested in triplicate in five independent experiments. Firefly luciferase activity was normalized to renilla luciferase activity for each transfected well. The normalized data were expressed as changes relative to the data of the cells transfected with 100 nM miR-181a-5p mimic, SCR was taken as 100%.

Statistical Analysis
Comparisons between groups were performed by the unpaired t-test. Data are given as mean 6 SD. Linear regression tests were performed; b regression coefficient (r) and r 2 were calculated. Results were considered statistically significant for p,0.05. Analyses were carried out using Statistical Package for Social Science (version 21.0; SPSS, Chicago, IL).   Figure S2 miRNA inhibitor transfection efficiency. miRCURY LNA microRNA Inhibitor Negative Control (100 nM) labeled with fluorescein (Exiqon, Vadbaek, Denmark) were transfected into PLC/PRF/5 cells with siPORTTM NeoFXTM (Life TechnologiesTM, Madrid, Spain), following manufacturer's instructions. After 6 hours transfection, cells were harvested and washed with PBS. Flow cytometry was performed using a BD FACSCalibur flow cytometer (BD Biosciences, Madrid, Spain) and samples were run through the flow cytometer until 2,000 events were collected. The mean 6 SD of transfection efficiency for three replicates was 92.1%60.3%. X-axis represents the intensity of fluorescence for FL1 channel in log scale and Yaxis the numbers of cells. We defined transfection efficiency as a percentage of cells positive for FL1 (dotted lines), taken as background signal the non-LNA transfected cells signal (solid lines). (TIF)