Emerging role of circANRIL expression in peripheral blood mononuclear cells as a promising liquid biopsy diagnostic biomarker for coronary artery disease

Background: Long-non-coding RNAs (lncRNAs) are increasingly recognized to play major roles in coronary artery disease (CAD). ANRIL, a well-known lncRNA is located on the 9p21.3 locus which is the rst reported CAD risk locus. This study aimed to evaluate the association of ANRIL and its transcript variants expressions with susceptibility to CAD as well as to investigate the impact of 9p21.3 locus single nucleotid polymorphisms (SNPs) on their expressions in adipose tissues and peripheral blood mononuclear cells (PBMCs) of Tanzanian study cohort. Methods: Expressions of ANRIL, and its variants; circANRIL (hsa_circ_0008574), NR003529, EU741058 and DQ485454 were detected by qRT-PCR in visceral adipose tissues (VAT) (epicardial adipose tissue (EAT) and mediastinal adipose tissue(MAT)), subcutaneous adipose tissue (SAT) and PBMCs of patients undergoing coronary artery bypass grafting due to CAD and non-CAD patients undergoing heart valve surgery. Five SNPs encompassing the 9p21.3 risk locus were genotyped in a cohort of 200 CAD patients an 220 non-CAD patients using qRT-PCR and their impact on gene expressions were further evaluated. Results: ANRIL expression was signicantly up-regulated, while the expression of circANRIL signicantly down-regulated in CAD patients compared to non-CAD patients. Also, increased circANRIL expression was determined to be signicantly inversely associated with the severity of CAD. Moreover, rs10757278 and rs10811656 polymorphisms were signicantly associated with expressions of ANRIL, EU741058, NR003529 and circANRIL in VAT and PBMCs. Decreased circANRIL levels were correlated with aggressive clinical characteristics. Moreover, the random forest model was evaluated to feature importance of CAD prediction. Accordingly, the receiver operating characteristics(ROC) curve analysis was evaluated and the result suggested that circANRIL has high diagnostic accuracy. The area under the ROC curve (AUC) was 0.9808, the optimal cut ‐ off value was 0.33, with a sensitivity of 1.0 and a specicity of 0.88. Conclusion: We report the rst data demonstating the presence of, ANRIL and its transcript variants expressions in adipose tissues and PBMCs of CAD patients and the strong association between ANRIL expression and CAD. circANRIL having a synergetic effect with ANRIL plays a protective role in CAD pathogenesis. Therefore, altered circANRIL expression may become a potential diagnostic biomarker for early CAD diagnosis in Tanzania.


Introduction
Coronary artery disease (CAD) is a complex disease and it is the most common cardiovascular condition. CAD is characterized by plaque build-up of fat and cholesterol in the arteries [1]. These plaques, when they are unstable or become obstructive, lead to CAD, the clinical manifestations of atherosclerosis [2]. The risk of developing CAD is related to interplay of genetic, lifestyle and environmental factors, their effects are demonstrated to be mainly mediated through cardiovascular risk factors.
Obesity is known to be the most prevalent risk factor for CAD [3]. Regional body adipose tissue distribution, rather than total body adiposity is a well-known marker for cardiovascular risk [4]. Especially, visceral adipose tissue (VAT), including epicardial adipose tissues (EAT) and mediastinal (MAT), has been studied in the last decade for their relationship with CAD. EAT is the most studied cardiac adipose tissue and it was gained increasing attention due to its location. EAT is located between the visceral pericardium and the heart, also it is in contact with the coronary vessels. Recent studies showed that EAT has paracrine or endocrine activities and it can secrete in ammatory factors, such as TNF-alpha, IL-6, and adipocytokines and it has the ability to release and uptake free fatty acids, which plays a vital role in the development for CAD [5,6]. Moreover, studies showed that the volume of EAT was demonstrated to be associated with the incidence of CAD [7][8][9][10]. Another cardiac adipose tissue is MAT, it is not close to the coronary vessels, but it is situated outside the pericardium that encloses the heart. Recently, the relationship between MAT and the development of CAD has been demonstrated and a study reported that the increased volume of MAT was associated with an enhanced CAD risk pro le [12]. Also, the volume of MAT was positively correlated with plasma triacylglycerol and C-reactive protein which they are wellknown risk factors and have a role in progression of the disease [13]. Furthermore, the fat depot's associated diseases such as CAD are related with excessive accumulation of subcutaneous adipose tissue (SAT), also an increase in the number of adipocytes in the SAT is accompanied by the presence of metabolic disorders [14].
As with most complex diseases, genetics play an important role for developing CAD. In recent genomewide association studies (GWAS) have identi ed a new susceptibility region located on human chromosome 9p21.3 (Chr9p21.3) containing multiple single nucleotide polymorphisms (SNPs) associated with CAD. [15][16][17][18]. The nearest genes, approximately 100 kilobases (kb) away from the core CAD region, are two tumor suppressor genes (cyclin-dependent kinase inhibitors) CDKN2A and CDKN2B that play an important role in cell cycle regulation, apoptosis, senescence, aging, and in ammation which are processes strongly involved in atherogenesis. From a functional perspective CDKN2A and CDKN2B are potentially functional candidates genes to implicated in the pathogenesis of atherosclerosis [19][20][21].
Antisense non-coding RNA in the INK4 locus (ANRIL, also known as CDKN2BAS-1) is another gene located at the same locus gene [22]. ANRIL (ENSG00000240498) spans a region almost 126.3 kb and ANRIL overlaps at its 5' end with CDKN2B and it has a role in the epigenetic regulation of the expression of adjacent protein-coding genes, including MTAP, CDKN2A and CDKN2B through multiple mechanisms, such as RNA interference, gene silencing, chromatin remodeling, or DNA methylation [23]. Recently, it was revealed that the risk locus could be responsible for regulating ANRIL expression to a certain extent, and also, 9p21.3 risk locus could have an impact CDKN2A and CDKN2B expression through ANRIL expression [24,25]. Another study demonstrated that deletion of the targeted 9p21.3 risk locus of orthologous ANRIL interval in mice reduced expression of CDKN2A and CDKN2B in the heart and lead to excessive proliferation of vascular cells that contribute development of atherosclerosis [26].
A few investigations suggested that transcript variants of ANRIL, not only in linear transcript variants but also in circular transcript variant, could also be responsible for the development of CAD [27][28][29][30]. Most studied linear transcript variants of ANRIL with CAD are as follows; the longest ANRIL transcript variant: NR003529 (3857 bp, ENST00000428597.6), and 2 shorter variants of ANRIL; DQ485454 (2659 bp, ENST00000580576.6), and EU741058 (688 bp, exons 1, 5 to 7, and 13, ENST00000455933.7). Recently, ANRIL transcripts (EU741058 and NR003529) were found to be signi cantly increased in peripheral blood mononuclear cells and atherosclerotic plaque tissue from patients of CAD compared to healthy individuals. Moreover, ANRIL expression was shown to be associated with CAD severity and with 9p21.3 risk genotypes [26]. Another study showed that subjects homozygous for the risk alleles of 9p21.3 exhibited increased expression levels of ANRIL short transcripts and decreased long variant expression levels, in comparison with healthy individuals carrying two copies of reference alleles [31].
Newly discovered circular transcript variant of ANRIL, circular antisense non-coding RNA in the INK4 locus (circANRIL or hsa_circ_0008574), consisted of exons 5, 6 and 7, where exon 7 was non-canonically spliced to exon 5. Many researchers have focused considerable attention on circular RNAs (circRNAs) in recent years due to there are evidence circRNAs are emerging that they may have importance in the development of tissues or organs and also in disease pathogenesis [32].
The circRNAs have different structures from their linear counterparts, circRNAs have a unique structure of covalently closed continuous loop lacking 5'cap and 3' poly-adenylated tails [33]. These differences make circRNAs more stable from their linear counterparts and they gain resistance to degradation by exonuclease and RNAses. Therefore, circRNAs are proposed as a new generation of predictive biomarkers and potential therapeutic targets for many diseases. The circANRIL was found in many different cell lines and in many primary cell types, including vascular smooth muscle cells, macrophages, heart, and vascular tissue [29,30]. Burd et al. found that circANRIL expression is associated with atherosclerotic risk. In addition, they also reported that circANRIL is more stable than linear ANRIL [30]. Another study by Holdt et al. speculated that circANRIL is a prototype of a circRNA regulating ribosome biogenesis and conferring atheroprotection, and in the same study they suggested that circANRIL remains a potential therapeutic target for the treatment of atherosclerosis [29]. Moreover, there is evidence the circANRIL isoform (exons 5-7) has a synergistic effect and play a more effective and protective role of CAD by controlling rRNA maturation and modulating pathways of atherogenesis [24].
Taken together observations from all studies done so far pointed out the importance of ANRIL and/or transcript variants of ANRIL in the mechanism mediating the 9p21.3 association, and suggested that ANRIL could be a possible candidate gene of CAD at the 9p21.3 risk locus.
Thus, the aim of the present study was rst to assess whether the adipose tissue and peripheral blood mononuclear cells (PBMCs) expressions of ANRIL and ANRIL transcript variants were associated with CAD susceptibility, and to investigate the impact of 9p21.3 locus variations on the expression of ANRIL and ANRIL transcript variants in PBMCs, in adipose tissues which play a role in the development of CAD (VAT (EAT and MAT) and SAT) in Tanzanian CAD patients.

Materials And Methods
Subjects: This study was performed at Jakaya Kikwete Cardiac Institute, Dar es Salaam, Tanzania. Participants enrolled in the study were selected among patients admitted to the cardiology department outpatient clinic for symptoms of angina, dyspnea and chest discomfort at the time of diagnosis, and patients admitted to the cardiology department for operation. A total of 420 subjects including 200 CAD patients (175 patients with CAD from outpatient clinic and 25 patients undergoing coronary artery bypass grafting (CABG) due to CAD) and 220 non-CAD patients (195 patients without CAD from outpatient clinic and 25 patients undergoing heart valve operation) were enrolled to invesitgate the association between the polymorphisms located on the 9p21.3 CAD risk locus and CAD susceptibility. Then 25 CAD patients undergoing CABG and 25 patients undergoing heart valve operation were included to detect ANRIL and its transcript variants expressions in adipose tissues and PBMCs. CAD was de ned as ≥ 50% luminal narrowing in at least one coronary artery from outpatient clinic and it was de ned as ≥ 50 % luminal narrowing in at least two coronary artery from cardiology department surgery clinic. CAD patients from cardiology department surgery clinic underwent bypass surgery due to at least two vessel disease according to the number of signi cantly affected vessels using the Coronary Artery Surgery Study classi cation [34]. Only the patients with normal coronary angiogram were included in the non-CAD patients. All subjects enrolled in this study were of Tanzanian origin.
Body mass index (BMI) was calculated as a ratio of mass in kilogram divided by the square of height in meters. Diagnosis of hypertension was de ned as resting blood pressure ≥ 140/90 mm Hg or current use of antihypertensive therapy. Diagnosis of diabetes was de ned either by 1999 World Health Organization (WHO) criteria 11 or self-report of being previously diagnosed as diabetic. Diagnosis of dyslipidemia was de ned as low-density lipoprotein (LDL) cholesterol ≥3 mmol/l or the use of hypolipidemic agents.
A detailed information of subjects on demographics, medical history, current medication and CAD risk factors such as diabetes, hypertension, smoking etc. were completed through personal interviews. Oral antidiabetic and lipid lowering drugs which may interfere with gene expression were stopped three days before the operation and sample collections. PBMCs and adipose tissue samples were obtained after overnight (12 hours) fasting. Fasting venous blood samples were centrifuged at 1500xg at 25°C for 10 minutes promptly and separated serum and plasma were stored at −80°C for further analyses.
Biochemical measurements: Serum total cholesterol (TC) and high-density lipoprotein cholesterol (HDL) and triglycerides (TG) were measured by routine enzymatic endpoint methods (Analyzer A15 Biosystems, Philippines). Fasting glucose was determined using the enzymatic reference method with glucose oxidase. Very low density lipoprotein cholesterol (VLDL) and LDL were calculated in keeping with Friedewald´s Formula [35].
Tissue biopsies: Approximately 30-50 mg of EAT, MAT and SAT were collected during operation in the form of biopsy. EAT samples were collected from the fat surrounding to the heart, within the pericardium and MAT samples were collected from the fat within the mediastinum, outside the pericardial sac. SAT samples were collected from retrosternal region. All adipose tissues and the tissue samples were then immediately frozen in liquid nitrogen and stored at −80°C prior to total RNA preparation.
DNA Extraction: Genomic DNA was obtained from peripheral blood leukocytes of all participants by the use of MagnaPure DNA Isolation robot (Roche, Germany) and DNA quantity was determined by NanoDrop™ 1000 Spectrophotometer (Thermo Scienti c, Wilmington, Delaware USA). The extracted DNA was stored at -80ºC and used later for the downstream analysis.
Genotyping: The SNPs rs10757274, rs2383207, rs2383206, rs10811656 and rs10757278 in all participants were detected with the use of LightSNiP typing assays (TIB MolBiol, Berlin, Germany), employing quantitative real-time polymerase chain reaction (QRT-PCR) ampli cations with melting curve analysis. The reactions were performed on a LightCycler®480 II Real-Time PCR system (Roche-Germany), following the recommendations of the manufacturer.
Total RNA extraction and quantitation: 30 mg of EAT, MAT and SAT samples were obtained from the subgroup including 25 CAD patients undergoing CABG operation and 25 non-CAD patients undergoing heart valve operation. Total RNA from EAT, MAT and SAT were extracted with the Trizol ™ Reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. RNA quantity was determined by NanoDrop™ 1000 Spectrophotometer (Thermo Scienti c, Wilmington, Delaware USA). After 1 μg of RNA was reverse transcribed in 20 μl total volume using OligodT as primers and Revert Aid First Strand cDNA Synthesis Kit (Fermentas, Canada). Following cDNA synthesis, the quality of cDNA was assessed by polymerase chain reaction (PCR) ampli cation of the housekeeping gene β-Globin. All cDNA samples were stored at −20 °C prior to gene expression analysis. Moreover, PBMCs were isolated from 5 ml anticoagulated peripheral blood, and stored at -80C until processed. Total RNA in PBMCs was extracted with Trizol ™ Reagent (Invitrogen, Carlsbad, CA, USA) and the concentration of RNA was quanti ed using NanoDrop™ 1000 Spectrophotometer (Thermo Scienti c, Wilmington, Delaware USA) and was reverse-transcribed into cDNA as described above.
Gene expression analysis by quantitative real-time PCR (QRT-PCR): The expression levels of ANRIL, circANRIL, NR003529, EU741058 and DQ485454 were detected by qRT-PCR using a LightCycler®480 II Real-Time PCR system (Roche, Germany). Ten-fold dilutions of cDNA synthesized from total RNA were used. All samples were ampli ed in duplicate and the mean values were obtained for further calculations. Primers and probes designed using the soft-ware Universal Probe Library and primers synthesized by IDT (Skokie-USA) and probes synthesized by Roche Applied Science (Roche-Germany).
Data analysis: Statistical analysis was performed using SPSS software (Statistical Package for the Social Sciences 25.0, SPSS Inc, Chicago, IL, USA). Quantitative variables were expressed as mean ± standard deviation (S.D.), and qualitative variables were expressed as percentages. The allelic frequencies of these polymorphisms between the non-CAD and CAD patients and the demographic characteristics between the groups were estimated for categorical variables were compared using Chi square (χ 2 ). Hardy-Weinberg equilibrium (HWE) was assessed by Fischer's exact test. The gene expression data were obtained as Cycle Threshold (CT) values (CT = cycle number at which logarithmic PCR plots cross a calculated threshold line). The expression of each gene was compared between depots using the 2 ΔΔCT method (ΔΔCT = CT of the target gene-CT of the housekeeping gene), respectively. The differences of normal and non-normal distributed continuous variables were compared using Student's t test and Mann-Whitney U test respectively. To evaluate differences between groups, the data were undergo log transformation to satisfy ANOVA criteria and then subjected to one-way ANOVA with Tukey's post-hoc analysis. The correlations between the gene expression and risk factors of CAD were evaluated with the Spearman correlation test. The statistical software package MedCalc Statistical Software (version 16.2, Ostend, Belgium) was used for multiple logistic regression analysis and the receiver operating characteristic (ROC) analysis to examine expressions of target genes in the PBMCs can be used as a biomarker for CAD. A logistic regression classi er implemented in the scikit-learn library [36] was used to perform leave-one-out cross-validation and to calculate the p-value using the method by DeLong et al. [37]. Normalized coe cient magnitudes was used as a means for calculating feature importances. Figures were generated in Python, using the matplotlib library [38]. Statistical signi cance was taken as p<0.05.

Results
Clinical and anthropometric characteristics of the study population: The clinical and anthropometric characteristics of the study group were presented in Table 1. There was not signi cant difference between the groups in terms of age. When compared the lipid pro les, serum TG, TC, LDL were higher in the all CAD patients than in the non-CAD patients (p<0.05, respectively), whereas serum HDL levels were signi cantly higher among non-CAD patients (p<0.05). The average fasting plasma glucose of the CAD patients were signi cantly higher than that of non-CAD patients (p<0.05). CAD patients had higher levels of systolic blood pressure and diastolic blood pressure. There was also a higher prevalence of obesity, hypertension, diabetes, hyperlipidemia and smokers in CAD patients compared to non-CAD patients (p<0.05). In addition, male gender was predominant in CAD patients (76%). Hypertension was the more commonly associated clinical condition in CAD patients (92%), while the least common one was determined to be the hyperlipidemia (56%). The further data analysis demonstrated that male gender, obesity, hypertension, diabetes mellitus, smoking and hyperlipidemia were the predominating risk factors for CAD in Tanzanian population. .67 95% CI= 2.14-14.99, respectively) in CAD patients compared to non-CAD patients of the subgroup ( Table 2). The analyses of allelic and genotypic showed that the rs2383207 SNP is not associated CAD risk in both groups.
ANRIL and ANRIL splice variants expression levels in adipose tissues and PBMCs from CAD patients and non-CAD patients: For expression analysis, we included 25 CAD patients undergoing CABG and 25 non-CAD patients undergoing valve replacement. The expression levels of ANRIL, and ANRIL transcripts; circANRIL, NR003529, EU741058 and DQ485454 were studied in EAT, MAT, SAT and PBMCs (Figure 1). ANRIL expression levels were signi cantly up-regulated in PBMCs of the CAD patients compared to non-CAD patients (fold change=1.6, p<0.001). Although ANRIL expression levels in EAT, MAT and SAT were found increased in CAD patients compared to non-CAD patients, the differences were not signi cant ( Figure 1A).
circANRIL was signi cantly down-regulated in PBMCs of CAD patients compared to non-CAD patients (fold change=5.3, p<0.001) ( Figure 1E) while no signi cant differences were determined in EAT, MAT and SAT. The associations of ANRIL and circANRIL expressions with the CAD severity (double stenotic vessels disease (n=13) versus triple stenotic vessels disease (n=12)) were also evaluated. A statistically signi cant difference in the expression levels of circANRIL was determined between the two groups (p<0.05) (Figure 2) but no statistically signi cant difference was observed for ANRIL expression (p>0.05).
Moreover, the expression levels of NR003529, EU741058 and DQ485454 in EAT, MAT, SAT and in PBMCs showed no statistical differences between CAD patients and non-CAD patients (p>0.05, respectively). The expression level of EU741058 in PBMCs was down-regulated in CAD patients compared to non-CAD (fold change≈0.8) but the difference was not signi cant. ANRIL and circANRIL were co-regulated with most of the risk factors of CAD such as lipid levels, blood pressure and glucose levels. These positive and negative correlation results with the risk factors, which mentioned above, suggested that ANRIL and circANRIL expressions regulate the risk factors leading to the CAD development and ANRIL and circANRIL may serve as indicator genes in CAD patients.
Future importance of the variables: Machine learning (ML) is a highly effective method for disease prediction using machine learning techniques. They are able to capture the complex interactions between predictors and outcomes during the data process, and they would provide a new and novel discernment towards the disease [39]. Recent studies showed that random forest (RF) is the most e cient algorithm for the prediction of CAD than the other algorithms and it gives consistently better accuracy of the prediction system [40][41][42]. We measured the importance of different features for the risk factors of CAD together with the expression levels of ANRIL and its transcript variants by the mean decrease impurity (Gini importance) of all decision trees in a tuned RF model. The importance of included variables obtained from the tuned RF model is presented in Figure 3. As expected, age, systolic BP, BMI and smoking were among the top risk factors. In addition we observed that expression levels of ANRIL and circANRIL in PBMCs were also among the top risk factors. To Finally, ROC analyses suggested that the detection of ANRIL expression and circANRIL expression together with risk factors of CAD exhibited a higher diagnostic performance compared with the detection of risk factors only. This result implied that the combination of ANRIL and circANRIL expressions in PBMCs has a great potential to be sensitive, and reliable biomarker possibly having a higher diagnostic value for CAD.

Discussion
While a big part of the human genome has already been transcribed, only ≈2% of the genome appears coding for protein, the rest being deemed "non-coding." Recent studies demonstrated that the human genome is prevalently transcribed and produced many thousands of non-coding RNAs (ncRNAs) such as microRNAs, small interfering RNAs, and various classes of long ncRNAs. A lot of ncRNAs play a vital role in cellular processes and they work with other biomolecules to modulate in gene regulatory networks and identi cation of the role of ncRNAs in the disease process could increase our understanding of the pathogenesis of complex diseases [43,44]. Moreover, ncRNAs have been considered within the molecular mechanisms affecting different diseases, particularly age-related complex disorders. Accumulating studies indicate that several long non-coding RNAs (lncRNAs) which are a member of the ncRNAs play important roles as key regulators of gene regulation in the development of cardiovascular diseases [45,46].
ANRIL is one of the most important antisense transcripts lncRNA, receiving extensive attention as potentially novel biological regulator. Many roles have been attributed to ANRIL including DNA damage response, DNA repair, epigenetic regulation of CDKN2A/B, controlling the cell cycle checkpoints, apoptosis and RNA splicing etc [47]. A growing number of recent studies con rm that ANRIL plays pivotal roles in diverse physiological and pathological processes in CAD [22,[24][25][26][27][28][29][30][31]. Moreover, the polymorhisms on 9p21.3 region are responsible for regulating transcript variants of ANRIL and expression of ANRIL in CAD [47]. The SNPs rs10757278 and rs10811656 are located in one of ANRIL gene enhancers binding site for STAT1. The risk allels of rs10757278/G and rs10811656/T disrupt a binding site for STAT1 and increases the expression of ANRIL [47].
We herein aimed to ascertain whether the ANRIL and ANRIL transcript variants were involved in the manifestation of CAD susceptibility and ANRIL transcripts are associated with the 9p21.3 CAD risk in the Tanzanian patients. The present study was reported for the rst time the expression of ANRIL and different transcript variants of ANRIL in PBMCs and in various adipose tissues.
Here we reported that differential expressions of ANRIL and its transcript variants in the CAD patients compared to non-CAD patients. ANRIL expression levels were signi cantly increased in PBMCs of CAD patients, but no statistically signi cant expression was observed in adipose tissues. There was also no statistical difference in terms of the expression levels of NR003529, EU741058 and DQ485454 in EAT, MAT, SAT and also in PBMCs between CAD and non-CAD patients. On the other hand, we discovered as the main ndings of our study; circANRIL expression was increased in PBMCs of CAD patients compared with non-CAD patients, and it well-discriminated CAD patients from non-CAD patients, and circANRIL level was associated with increased CAD severity.
Our study demonstrated that up-regulated ANRIL expression could possibly be involved in CAD pathogenesis. This is in concordence with other studies which investigated the association of ANRIL expression and CAD in different populations [48,49]. Moreover, it was demonstrated that increased expression of ANRIL lead to increased in ammation, CAD risk, severity and poor prognosis of CAD [50][51][52]. ANRIL expression was also showed to be associated with in ammation cytokines; monocyte chemoattractant protein-1 and interleukin-10 [52] and the knockdown of ANRIL signi cantly promoted cell proliferation and tubule formation and inhibited in ammatory activation and apoptosis of human umbilical vein endothelial cell [52]. In contrast, a study from Yang et al. reported no signi cant association between CAD and ANRIL expression in the Chinese Han population [53].
Furthermore, we observed a signi cant down-regulation of circANRIL expression in CAD patients compared to non-CAD patients and we suggested that circANRIL may play an atheroprotective role in CAD pathogenesis, which would indicate a synergetic effect between circANRIL and ANRIL. Holdt et al. also suggested that, circANRIL has a protective role against atherosclerosis [29]. Moreover, recent studies showed that increased circANRIL expression in human primary smooth muscle cells and vascular endothelial cells in a rat model of CAD could induce apoptosis and inhibit the proliferation and thereby increased circANRIL expression could decreased to susceptibility of CAD [54,55].
But we failed to detect the same signi cance of ANRIL and circANRIL expression levels in adipose tissues. One potential explanation of this result would be that ANRIL and circANRIL could be differentially expressed in adipose tissues and PBMCs.
We also found decreased expression levels of EU741058 in PBMCs of CAD patients compared to non-CAD patients but the differences was a borderline statistical signi cance. Previously, a few studies reported that EU741058 variant down-regulated in CAD patients and these results in agreement with our nding [53,56].
Further analysis revealed that rs10757278 and rs10811656 were altered the expression levels of ANRIL and ANRIL transcript variants not only in PBMCs but also in adipose tissues. This result emphasized that rs10757278 and rs10811656 could controlling the expression levels of ANRIL and ANRIL transcript variants.
The circANRIL expression levels in PBMCs were signi cantly down-regulated in the CAD risk genotype carriers of rs10757278 (GA and GG) and rs10811656 (CT and TT) compared to wild type carriers. Also, the risk genotypes carriers of rs10757278 (GA and GG) and rs10811656 (CT and TT) had signi cantly higher ANRIL expression in EAT, MAT and in PBMCs of CAD patients. A recent study by Liu et al. showed that the expression of ANRIL signi cantly decreased in rs10757278 risk allele carriers in T cells of 170 healthy individuals [57]. The potential explanations of the diverse results between the research by Liu et al., and our study might be the samples used were T cells from healthy individuals and in our research the samples were EAT, MAT and PBMCs from patients suffering from CAD [57].
Also, the expression levels of NR003529 and EU741058 in EAT, MAT and in PBMCs were statistically signi cantly higher in the CAD patients carrying the risk genotype of rs10757278 and rs10811656 compared to wild type carriers, but the expression of DQ485454 remained unaffected. This nding indicated that NR003529 and EU741058 were abundantly transcribed in CAD risk genotype carriers leading to the development of CAD.
It is well known that EAT is an important component of VAT due to its contiguity to the coronary arteries and the effects of endocrine and paracrine activity secreting pro-in ammatory and anti-in ammatory cytokines and chemokines, therefore it has been suggested to in uence coronary atherosclerosis development [58,59]. In recent years MAT also has gained increasing attention on account of the contributions of development of atherosclerosis [60, 61]. Our results also showed that the clinical importance of EAT and also MAT in CAD.
Another nding in our study was the increased expression levels of circANRIL in PBMCs which was also associated with less severity of CAD.
The circANRIL expression was statistically signi cantly decreased in triple stenotic vessels disease compared to double stenotic vessels disease (2 fold). A previous study reported that EU741058 and NR00359 expressions were pre-dominantly associated with severity of CAD [27]. Although, we could not nd the association between EU741058 and NR00359 expression and CAD severity but we successfully showed that the circANRIL expression was directly associated with severity of CAD.
In clinical practice, the increased levels of glucose, TC, TG, LDL and BMI, decreased HDL levels and also hypertension were always being attributed as major risk factors for CAD. In our study, we observed that these risk factors of CAD correlated with ANRIL and circANRIL expression levels in the blood. This coregulation suggested that ANRIL and circANRIL regulate the risk factors leading to the CAD development and they may well serve as early disease indicators in CAD patients.
Although the CAD mortality rate is high, the chance of survival is higher if the diagnosis is made early enough in an accurate and e cient way. Therefore, the experts attempt to nd a way for the accurate and timely prediction of CAD by using new statistical techniques, such as data mining which would help to identify the risk models of the disease, to recognize the disease patterns and the in uencing factors. The feature importance selection methods give us an opportunity to increase the accuracy of the prediction of disease compared to traditional methods of classi cation. Especially, the combination of the feature importance selection algorithm such as RF, which has better accuracy for CAD prediction than other algorithms, is necessary for the e cient prediction of CAD [42,62]. We examined the importance of different features for the risk factors of CAD together with the expression levels of ANRIL and its transcript variants and we observed that age, systolic BP, BMI, and smoking were among the top risk factors as we expected. In addition, the expression levels of ANRIL and circANRIL in PBMCs were also among the top risk factors.
The ROC analysis was done accordingly and the analyses showed that a good predictive value of circANRIL (exon 5-7) expression together with risk factors of CAD exhibited a higher diagnostic performance compared to CAD risk factors only. The circANRIL expression levels in PBMCs may be a new and non-invasive diagnostic tool for the diagnosis of CAD. As a screening tool in clinical practice, circANRIL has potential diagnostic value and is worthy of clinical promotion [63]. Moreover, the circANRIL, a circular transcript variant of ANRIL, binds and negatively regulates Pescadillo Ribosomal Biogenesis Factor 1 (PES1) protein in the PeBoW complex known as an essential complex to ribosome biogenesis in macrophages and vascular smooth muscle cells. And this negatively regulation induces p53 activation and nucleolar stress, resulting in the inhibition of proliferation and induction of p53 in proliferating cells [29] ( Figure 5).
Taken together, our results emphasized that, the remarkable changes of ANRIL and circANRIL expression levels could be in uenced by the progression of atherosclerosis in CAD by directly or indirectly ways. ANRIL and circANRIL could be a feature as potential biomarkers as well as therapeutic targets and also they might be a useful disease marker.

Conclusions
In conclusion, we genetically analyzed the expression levels of ANRIL and its transcript variants in adipose tissues and in PBMCs, and we further analyzed the associations between expression levels of ANRIL and its transcript variants and the variations of SNPs located on 9p21.3 region. To the best of our knowledge, this is the rst study showing the expression levels of ANRIL and its transcript variants in VAT in CAD patients, and suggesting presence of an association between the expression levels of ANRIL and its transcript variants in VAT and CAD susceptibility.
Our results revealed that ANRIL expression levels in PBMCs were signi cantly increased while circANRIL expression levels were signi cantly decreased in CAD patients when compared to non-CAD patients. Additionally, circANRIL expression was found to be related with the severity of CAD. The rs10757278 and rs10811656 SNPs were also determined to be associated with not only ANRIL but also transcript variants of ANRIL (EU741058, NR003529 and circANRIL). Not last, the RF model and ROC analyses showed that measuring the expression levels of circANRIL (exons 5 to 7) might offer a well reliable and sensitive prognostic value and allow for early detection and better monitor treatment response or disease recurrence for CAD. Taken together, our ndings can provide a new liquid biomarker to be used together with the clinical decision-making and improve the early diagnosis and treatment of cardiovascular diseases. Figure 3 Feature importance analysis of included variables obtained from a tuned random forest model in Tanzanian CAD patients. BMI: body mass index, SBP: systolic blood pressure, DBP: diastolic blood pressure.