Trimethylamine N-oxide and the reverse cholesterol transport in cardiovascular disease: a cross-sectional study

The early atherosclerotic lesions develop by the accumulation of arterial foam cells derived mainly from cholesterol-loaded macrophages. Therefore, cholesterol and cholesteryl ester transfer protein (CETP) have been considered as causative in atherosclerosis. Moreover, recent studies indicate the role of trimethylamine N-oxide (TMAO) in development of cardiovascular disease (CVD). The current study aimed to investigate the association between TMAO and CETP polymorphisms (rs12720922 and rs247616), previously identified as a genetic determinant of circulating CETP, in a population of coronary artery disease (CAD) patients (n = 394) and control subjects (n = 153). We also considered age, sex, trimethylamine (TMA) levels and glomerular filtration rate (GFR) as other factors that can potentially play a role in this complex picture. We found no association of TMAO with genetically determined CETP in a population of CAD patients and control subjects. Moreover, we noticed no differences between CAD patients and control subjects in plasma TMAO levels. On the contrary, lower levels of TMA in CAD patients respect to controls were observed. Our results indicated a significant correlation between GFR and TMAO, but not TMA. The debate whether TMAO can be a harmful, diagnostic or protective marker in CVD needs to be continued.

TMAO and TMA in CAD patients and controls. No differences were noted in row values of plasma TMA between controls 0.62 ± 0.13 μM (mean ± SD) and CAD patients 0.60 ± 0.11 μM (Fig. 1A). However, Generalized Linear Model (GLM) analysis, including adjustments for glomerular filtration rate (GFR), age, body mass index (BMI) and sex, identified a significant difference between the two groups for TMA (expected marginal means ± SD: controls = 0.63 ± 0.01 μM; CAD patients = 0.60 ± 0.01 μM; p = 0.004). TMAO was not significantly different between controls and CAD patients (Fig. 1B), regardless of the row values (p = 0.712) or in the analysis adjusted for the covariates (p = 0.251). www.nature.com/scientificreports/ Genotyping. Genotype and minor allele frequencies of the selected polymorphisms are reported in Table 2.
All the polymorphisms were in Hardy-Weinberg Equilibrium (HWE) (p > 0.05) and minor allele frequencies (MAF) at both rs12720922 and rs247616 SNPs were consistent with Northern Europe reference population data ( Table 2).
CETP SNPs are directly associated with HDL-cholesterol levels. Since most of the CAD patients were treated with statins (commonly used as primary or secondary prevention measurement), we relied on a Mendelian randomization-based approach to study the impact of CETP and HDL-cholesterol on TMA and TMAO. Despite the potential interference of statins treatment, rs12720922 and rs247616 CETP SNPs were significantly associated with HDL-cholesterol levels in the total population ( Supplementary Fig. S1 online). Conversely, these polymorphisms were not associated with LDL-cholesterol or total cholesterol levels. This evidence suggests that rs12720922 and rs247616 SNPs can selectively predict HDL-cholesterol even in presence of statin treatment. However, since the risk of unpredictable effects due to the statin treatment cannot be excluded (Supplementary Table S1 online), we confirmed the usage of the Mendelian randomization-based approach for the subsequent analysis and did not consider the raw data on lipid profile.
CETP SNPs are not directly associated with CAD. Chi-square analysis revealed that genotypes were not differently distributed among controls or CAD patients, thus neither CETP rs247616 (p = 0.426) nor rs12720922 (p = 0.488) appear to be directly associated with CVD considering a codominant model. Moreover, no associations were detected using additive models; similarly, no differences in the distribution of alleles between the two classes were detected for any of the analysed SNP ( Table 3).
Effects of different CETP genotypes on TMAO, TMA and TMAO/TMA. CETP rs12720922 genotype was associated with TMAO levels (p = 0.008) and TMAO/TMA ratio (p = 0.018) (GLM analysis; sex, age and GFR as covariates; Fig. 2); conversely, it was not linked to TMA levels (p = 0.159). Accordingly, the reces-  TMAO, TMA in CVD; CETP genetic background association. GLM analysis showed a different association between TMAO or TMAO/TMA levels and health status (controls vs CAD patients) depending on the rs247616 genotype. In particular, the rs247616-CC individuals belonging to the control group displayed lower TMAO levels than the carriers of the same genotype in the CAD group. On the other hand, T carriers, that had higher TMAO values in controls, exhibited reduced TMAO levels in the CAD group (P = 0.049) ( Supplementary  Fig. S2A online). This evidence preliminarily suggested that the increase of TMAO in CAD is typical of those individuals that carry the rs247616-CC risk genotype (associated to genetically determined higher CETP and lower HDL levels), but is not generalizable to the entire population. A similar effect was observed for TMAO/ TMA ratio, which was different in the control or CAD group depending on the rs247616 genotype (p = 0.046)  Haplotype association with CVD, TMAO and TMA. Analysis of haplotypes revealed that it was not possible to demonstrate a cumulative effect of the SNPs from data collected in this study. Indeed, distribution of haplotypes in CAD patients was not different in comparison to controls (p = 0.19) (Table 4). Moreover, there was not a significant association between haplotypes and TMAO (global haplotype association, p = 0.45) nor TMA levels (global haplotype association, p = 0.16) ( Table 5).

Discussion
In this study, we found no association between TMAO levels and genetically determined CETP in a population of CAD patients and control subjects. Moreover, we noticed no differences between CAD patients and control subjects in plasma TMAO levels.
In particular, we investigated two SNPs, rs247616 and rs12720922, as largely determining CETP concentration 22 . An increase in genetically determined serum CETP concentration has been previously associated with decreased total cholesterol concentration and HDL-cholesterol concentration 22 , with CETP as an important determinant of HDL-cholesterol, but not affecting LDL-cholesterol concentration and composition 23 . This evidence was essential in the design of this study since direct measurement of HDL-and LDL-cholesterol were not reliable markers in the recruited population, because most of the CAD patients were treated with statins (commonly used as primary or secondary prevention measurement). Results on CETP rs247616 genotyping were similar to those previously shown in the Polish population 24 . Despite the comparability in CETP rs247616 genotype and the higher number of subjects recruited, we were not able to observe significant differences on the rs247616 genotypes distribution between CAD patients and control groups. Similarly, no significant differences were observed for the rs12720922 genotype, revealing that the risk-alleles were not differently distributed between controls or CAD patients. Thus, we failed to find an association between the HDL-cholesterol increasing genotypes of CETP to CVD. It must be noted that genetic mechanisms raising plasma HDL-cholesterol do not decrease the risk of myocardial infarction 25 , and only SNPs affecting LDL-cholesterol levels or both, LDLcholesterol and HDL-cholesterol levels, influence CVD risk 26 .
Moreover, data collected in the current study did not support the hypothesis that TMAO is directly associated with CVD. We observed similar plasma TMAO levels in patients with confirmed angiographically CAD and control subjects with no medical history of CVD, and plasma TMAO concentration were coherent with www.nature.com/scientificreports/ values previously measured in the general population 27 . Moreover, no significant pure associations between the CETP genotypes and TMAO metabolism has been found. Nevertheless, some aspects of the CETP genotype can be mentioned. Firstly, higher TMAO levels have been measured in the rs12720922-AA carriers, which are the subjects with genetically elevated circulating CETP and lower HDL-cholesterol levels. On the contrary, rs12720922-G carriers displayed similar levels of TMAO in both groups. However, it must be noticed that the group of s12720922-AA carriers in CAD patients is limited to a very small number of subjects (n = 12), which is 3.0% of examined CAD population. Secondly, preliminary evidence suggested that the association between high TMAO and CAD is peculiar of the rs247616-CC risk genotype (which is associated to higher CETP and lower HDL levels), but is not generalizable to the entire population. Thus, the involvement of CETP in CAD seems to be more complex than initially hypothesized 24 , and the association between TMAO and CAD might be not as strong as previously suggested 28,29 . In fact, despite previously reported the pro-atherogenic effect of TMAO 13 , recent studies did not observe a positive correlation between plasma TMAO concentrations and atherosclerosis development 30,31 . Previous evidence suggested an important implication of HDL metabolism in modulating the association between TMAO and atherosclerosis. Firstly, since the production of TMAO is dependent on liver FMO3 15 , genetic variants of FMO3 have been implicated in a number of diseases 32 and TMA/FMO3/TMAO has been identified as a key pathway 16,33 . In particular, expression of FMO3 modifications in LDLR −/− mice alters circulating and hepatic lipid levels 16 . Moreover, knockdown of FMO3 reorganizes whole body cholesterol balance by regulation of reverse cholesterol transport 33 . Moreover, in humans, FMO3 is significantly associated with age, gender, and genotype 34 . Indeed, several cofounding factors that mediates the association between TMAO and atherosclerosis has been identified. We have not determined FMO3 genotype, but differences in TMA/TMAO ratio due to differences in the amount and activity of FMO3 might be present in our population 16,35 . For this reason, both age and gender were a priori selected as covariates in statistical analyses. Another aspect to consider is that CVD and kidney disease (KD) are closely interrelated 36 and diminished renal function is strongly associated with morbidity and mortality in heart failure patients 37 . In ApoE −/− mice model of atherosclerosis, the hypercholesterolemia led to early renal dysfunction that can progress into chronic KD 38 . In chronic KD, TMAO elimination from the body fails, causing the elevation of its plasma concentration 39 . Therefore, higher plasma TMAO in humans was suggested as a marker of kidney damage 40 . Since plasma TMAO has been inversely correlated with GFR 41 , some studies suggest that GFR can be a cofounder in this association [42][43][44] . Moreover, in the end-stage KD patients, not only TMAO but also plasma TMA is elevated 39 . Thus, we also added GFR as a covariate in the analysis investigating the relationship between TMA/TMAO levels and CVD, so we can exclude that GFR could be responsible for the observed results.
Finally, it is worthy of note that chronic, low-dose oral TMAO treatment showed a reduction in diastolic pressure and cardiac fibrosis in spontaneously hypertensive rats 45 . Since TMAO stabilize proteins against various environmental stress factors, including high hydrostatic pressure 46 , TMAO has been suggested as a result rather than a cause of CVD 29 . Thus, not TMAO, but TMA has been suggested as implicated in CVD 47 . In our results, marginally lower levels of TMA in CAD patients respect to controls were observed. Therefore, the microbial origin of TMA is of great interest. Indeed, a major role is played by the microbiome in regulating health and wellbeing 48 , and dysbiosis of the gut microbiota has been measured in stroke and transient ischemic attack patients whose blood TMAO levels were decreased 49 .
In conclusion, the studied polymorphisms had no direct roles in the development of CVD in the studied Polish population. Moreover, we observed no differences between CAD patients and control subjects in plasma TMAO levels, TMAO which can be affected by intra-individual variation 50 . The debate whether TMAO can be a harmful, diagnostic or protective marker in CVD 28,29,32 has to be continued.

Materials and methods
Participants. CAD patients were consecutively recruited in one hospital with angiographically confirmed CAD or with angina referred to elective or urgent coronary angiography as inclusion criteria. The diagnosis of ST-segment elevation myocardial infarction (STEMI) and non-ST-segment elevation myocardial infarction (NSTEMI) was established according to the Third Universal Definition of Myocardial Infarction, and unstable angina (UA) was diagnosed according to the 2015 ESC guidelines for the management of NSTE-ACS3 51,52 . Control subjects were recruited in the same region amongst the subjects without a self-reported medical history of CVD. The study was approved by the Regional Bioethical Committee (RBC) in Gdansk (KB-27/16 and KB 32-17). All methods were carried out in accordance with relevant guidelines and regulations approved by RBC. Informed consent was obtained from all subjects.
Samples collection. Venous blood samples were collected in EDTA-containing tubes. The plasma samples were prepared by centrifugation at 1300×g for 10 min at 18-25 °C, and were kept frozen at − 80 °C for later TMA and TMAO analysis.
TMA and TMAO analyses. Plasma TMA and TMAO were determined by the Ultra-Performance Liquid Chromatography (UHPLC) tandem mass spectrometry method, based on the methods described previously 53,54 . UHPLC separation was performer on an XBridge HILIC 3.5 μm (3.0 mm × 50 mm) column on a NEXERA Shimadzu UHPLC system coupled with QT4500 SCIEX. Trimethyl-d 9 -amine HCl (d 9 -TMA) was used as an internal standard. The 3 μM of d 9 -TMA working solution of internal standard (ISWS) was prepared in methanol/acetonitrile (15:85) and 0.1% formic acid (v/v). Calibration samples, QC and plasma samples were prepared by addition 100 μl of cold ISWS to 50 μl of each sample type. All samples were vortexed and kept on ice for 15 min for protein precipitation. Centrifuged samples ( Statistical analysis. The sample size was calculated through a power analysis performed by G*Power. The effect size of TMAO variation in CAD patients respect to controls was calculated from the study of Tang and colleagues 18 , which has been identified as a high-quality study in the meta-analysis from Qi and colleagues 55 . The calculated effect size is 1.158; thus, to have a power of 0.95, the minimum sample size is 34 subjects (see Supplementary Fig. S3 online). Power analysis has been performed using G*Power software 56 . The Shapiro-Wilk test was used for the analysis of the normality of data distribution. Spearman correlation, Chi-square test, Kruskal-Wallis test and Generalized Linear Model (GLM) were used to test correlations and significant differences among analysed variables. Hardy-Weinberg equilibrium was calculated for all the Single Nucleotide Polymorphisms (SNPs) analysed. The best fitting model of the association was determined using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) provided by SNPStats. The model with the lowest AIC and BIC values was considered the best fitting model. Haplotype frequencies estimation and global haplotype association were calculated using SNPstats 57 . If not differently specified, statistical analyses were performed using the SPSS package for Windows, v.20.0 (SPSS Inc, Chicago, IL).

Data availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.