ARTICLE doi:10.1038/nature09922 Gut flora metabolism of phosphatidylcholine promotes cardiovascular disease Zeneng Wang1,2, Elizabeth Klipfell1,2, Brian J. Bennett3, Robert Koeth1, Bruce S. Levison1,2, Brandon DuGar1, Ariel E. Feldstein1,2, Earl B. Britt1,2, Xiaoming Fu1,2, Yoon-Mi Chung1,2, Yuping Wu4, Phil Schauer5, Jonathan D. Smith1,6, Hooman Allayee7, W. H. Wilson Tang1,2,6, Joseph A. DiDonato1,2, Aldons J. Lusis3 & Stanley L. Hazen1,2,6 Metabolomics studies hold promise for the discovery of pathways linked to disease processes. Cardiovascular disease (CVD) represents the leading cause of death and morbidity worldwide. Here we used a metabolomics approach to generate unbiased small-molecule metabolic profiles in plasma that predict risk for CVD. Three metabolites of the dietary lipid phosphatidylcholine—choline, trimethylamine N-oxide (TMAO) and betaine—were identified and then shown to predict risk for CVD in an independent large clinical cohort. Dietary supplementation of mice with choline, TMAO or betaine promoted upregulation of multiple macrophage scavenger receptors linked to atherosclerosis, and supplementation with choline or TMAO promoted atherosclerosis. Studies using germ-free mice confirmed a critical role for dietary choline and gut flora in TMAO production, augmented macrophage cholesterol accumulation and foam cell formation. Suppression of intestinal microflora in atherosclerosis-prone mice inhibited dietary-choline-enhanced atherosclerosis. Genetic variations controlling expression of flavin monooxygenases, an enzymatic source of TMAO, segregated with atherosclerosis in hyperlipidaemic mice. Discovery of a relationship between gut-flora-dependent metabolism of dietary phosphatidylcholine and CVD pathogenesis provides opportunities for the development of new diagnostic tests and therapeutic approaches for atherosclerotic heart disease. The pathogenesis of CVD includes genetic and environmental factors. A known environmental risk factor for the development of CVD is a diet rich in lipids. A relationship between blood cholesterol and triglyceride levels and cardiovascular risk is well established. However, less is known about the role of the third major category of lipids, phospholipids, in atherosclerotic heart disease pathogenesis. Another potential yet controversial environmental factor in the development or progression of atherosclerotic heart disease is inflammation due to infectious agents. Some studies have reported associations between coronary disease and pathogens such as cytomegalovirus (CMV), Helicobactor pylori, and Chlamydia pneumoniae1–4. However, prospective randomized trials with antibiotics in humans have thus far failed to demonstrate cardiovascular benefit5–7 and studies with germfree hyperlipidaemic mice confirm that infectious agents are not necessary for murine atherosclerotic plaque development8. Although a definite cause-and-effect relationship between a bacterial or viral pathogen and atherosclerosis in humans has not yet been established, the prospect of a role for microbes in atherosclerosis susceptibility remains enticing. The intestinal microbiota (‘gut flora’), comprised of trillions of typically non-pathogenic commensal organisms, serve as a filter for our greatest environmental exposure—what we eat. Gut flora have an essential role, aiding in the digestion and absorption of many nutrients9. Animal studies have recently shown that intestinal microbial communities can influence the efficiency of harvesting energy from diet, and consequently influence susceptibility to obesity10. Metabolomics studies of inbred mouse strains have also recently shown that gut microbiota may have an active role in the development of complex metabolic abnormalities, such as susceptibility to insulin resistance and nonalcoholic fatty liver disease11. A link between gut-flora-dependent phospholipid metabolism and atherosclerosis risk through generation of pro-atherosclerotic metabolites has not yet been reported. Metabolomics studies identify markers of CVD In initial studies we sought to discover unbiased small-molecule metabolic profiles in plasma that predict increased risk for CVD. An initial ‘Learning Cohort’ was used comprising plasma from stable patients undergoing elective cardiac evaluation who subsequently experienced a heart attack (myocardial infarction), stroke or death over the ensuing three-year period versus age- and gender-matched subjects who did not. Liquid chromatography with on-line mass spectrometry (LC/ MS) analysis of plasma was performed to define analytes associated with cardiac risk as described in Methods. Of an initial 2,0001 analytes monitored, 40 met all acceptability criteria within the Learning Cohort. Subsequent studies within an independent ‘Validation Cohort’ led to identification of 18 analytes that met acceptability criteria in both Learning and Validation Cohorts (Fig. 1a, b, Supplementary Fig. 1a and Supplementary Table 1). The structural identity of the 18 small molecules in plasma, the levels of which track with cardiac risks, was not known, as the compounds were screened on the basis of retention time and mass-tocharge ratio (m/z) when analysed by LC/MS. Among the 18 analytes, those with m/z 76, 104 and 118 demonstrated significant (P , 0.001) correlations among one another, suggesting a potential relationship via a common biochemical pathway (Supplementary Fig. 1b). We therefore initially sought to structurally define these three analytes. Phosphatidylcholine metabolites are linked to CVD The candidate compound in plasma with an m/z of 76 associated with CVD risks was isolated and unambiguously identified as TMAO using 1 Department of Cell Biology, Cleveland Clinic, Cleveland, Ohio 44195, USA. 2Center for Cardiovascular Diagnostics and Prevention, Cleveland Clinic, Cleveland, Ohio 44195, USA. 3Department of Medicine/ Division of Cardiology, BH-307 Center for the Health Sciences, University of California, Los Angeles, California 90095, USA. 4Department of Mathematics, Cleveland State University, Cleveland, Ohio 44115, USA. 5Bariatric and Metabolic Institute, Cleveland Clinic, Cleveland, Ohio 44195, USA. 6Department of Cardiovascular Medicine, Cleveland Clinic, Cleveland, Ohio 44195, USA. 7Department of Preventive Medicine and Institute for Genetic Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California 90089, USA. 7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 5 7 ©2011 Macmillan Publishers Limited. All rights reserved RESEARCH ARTICLE P for trend <0.05 43 analytes 29 analytes 25 analytes 40 analytes 24 analytes 18 analytes (3) Structural identiication of analytes (4) Conirm clinical prognostic utility in independent prospective cohort (N = 1,876) OR (95% CI) 30 132.3 34.9 34.9 b Analytes presumably related via common pathway 20 10 1 0 50 100 150 200 m/z Figure 1 | Strategy for metabolomics studies to identify plasma analytes associated with cardiovascular risk. a, Overall schematic to identify plasma analytes associated with cardiac risk over the ensuing 3-year period. CVA, cerebrovascular accident; HPLC, high-performance liquid chromatography; MI, myocardial infarction. b, Odds ratio (OR) and 95% confidence intervals (CI) of incident (3-year) risk for MI, CVA or death of the 18 plasma analytes that met all selection criteria in both Learning and Validation Cohorts; odds ratio and 95% confidence intervals shown are for the highest versus lowest quartile for each analyte. Filled circles represent the analytes (m/z 5 76, 104, 118) focused on in this study. m/z, mass to charge ratio. Gut flora is needed to form TMAO from dietary PC Intestinal microflora have a role in TMAO formation from dietary free choline13. We therefore proposed that commensal organisms (gut flora) might also have an obligate role in TMAO formation from dietary PC. To test this, deuterated PC was synthesized whereby the choline-methyl groups were deuterium labelled (that is, d9-PC) and Betaine a CH 3 Choline CH 2 CH 3 [O 2 m/z = 104 d4-choline: (CH3)3+NCD2CD2OH Or al Gu ora d9-DPPC gavage 0.1 olin e TM 0 0 2 3 CH 3 FMO H [O] CH CH 0 4 Time (h) 1.6 3 OH 3 25 0.8 0.02 Choline TMAO 0 2 Time (h) 0 4 e in ol Ch TMAO 0 0 2 4 Time (h) Acquistion of microflora 0.04 ne tai Be Post-antibiotics Betaine 0 CH 50 m/z = 76 d4, m/z = 76 d9, m/z = 85 m/z = 60 d4, m/z = 60 d9, m/z = 69 Concentration (μM) Concentration (μM) Concentration (μM) Ch AO ly CH 3 Betaine 1.0 TMAO TMA CH 3 0.2 2.0 d9-DPPC i.p. m/z = 118 d4, m/z = 120 d9, m/z = 127 Suppresion of microflora Pre-antibiotics 2 3 on t fl CH CH ] d9-choline: (CD3)3+NCH2CH2OH c 3 N+ N+ 3 CH CH CH HO CH . i.p or al Or 3 N+ multinuclear nuclear magnetic resonance (NMR), multi-stage mass spectrometry (MSn), liquid chromatography with tandem mass spectrometry (LC/MS/MS) and gas chromatography with tandem mass spectrometry (GC/MS/MS) after multiple derivatization strategies (see Methods, Supplementary Figs 2a–d, and Supplementary Table 2). TMAO, an oxidation product of trimethylamine (TMA), is a relatively common metabolite of choline in animals12,13. Foods rich in the lipid phosphatidylcholine (PC, also called lecithin), which predominantly include eggs, milk, liver, red meat, poultry, shell fish and fish, are believed to be the major dietary sources for choline, and hence TMAO production14. Briefly, initial catabolism of choline and other trimethylamine-containing species (for example, betaine) by intestinal microbes forms the gas TMA13, which is efficiently absorbed and rapidly metabolized by at least one member of the hepatic flavin monooxygenase (FMO) family of enzymes, FMO3, to form TMAO15,16. Identification of the plasma analyte associated with CVD risk with an m/z of 76 as TMAO therefore indicated that the plasma analyte with an m/z of 104 might be choline. Further, these results also indicated that the plasma analyte with an m/z of 118 associated with CVD might be related to PC (choline) metabolism. To test the hypothesis that the plasma analytes with m/z 76 (TMAO), 104 and 118 might all be derived from the major dietary lipid PC, mice were fed egg-yolk PC (through oral gavage) and plasma levels of analytes over time were monitored. In both male and female mice, analytes with the same m/z (76, 104 and 118) and the same retention times as the corresponding analytes of interest observed in human plasma all increased after oral PC feeding (Supplementary Fig. 3a, b), strongly indicating that the m/z 104 analyte was choline, and the analyte at m/z 118 was derived from PC. Confirmation that the plasma analyte (m/z 104) associated with CVD risk was choline was achieved by b C 2.0 Conventionalized 0.2 TMAO Choline 1.0 0.1 Betaine 0 0 0 2 Concentration (μM) P for trend <0.05 Concentration (nM) 58 analytes HPLC-MS Adjusted –Log(P)>1.3 Concentration (μM) Concentration (μM) HPLC-MS Adjusted –Log(P)>1.3 MSn, LC/MS/MS and GC/MS/MS after multiple derivitization strategies (Supplementary Fig. 4a–d and Supplementary Table 3). We next studied the plasma analyte with m/z 118. We proposed that the analyte was either betaine or one of several potential methylated metabolites of choline (see Supplementary Fig. 5a for structures and strategy for discrimination among these isomers). To distinguish between these species, and explore a role for intestinal generation of the various metabolites, different isotopically labelled choline precursors were administered to mice either through an oral (gavage) or a parenteral (intraperitoneal, i.p.) route. The observed m/z of new isotopically labelled analytes at the appropriate retention times identified in plasma after these isotope tracer studies are summarized in Fig. 2a. Oral administration of non-labelled choline resulted in time-dependent increases in plasma levels of analytes with m/z 76, 104 and 118, consistent with TMAO, choline and either betaine or a methylated choline species (Supplementary Fig. 6a). Use of selectively deuterated choline species at either the trimethylamine moiety (d9 isotopomer) or the ethyl moiety (d4 isotopomer) unambiguously confirmed the m/z 118 analyte as betaine (Fig. 2a and Supplementary Fig. 6b). Further confirmation was acquired by observing the same retention time in LC/MS and an identical collision-induced dissociation (CID) mass spectrum (Supplementary Fig. 5b). Moreover, supplementation of PC or choline isotopomers via gavage or i.p. injection showed an absolute requirement for the oral route in TMAO production, whereas betaine production from PC or choline was formed via both oral and i.p. routes (Fig. 2 and Supplementary Fig. 7a). N+ Validation Cohort 25 cases (3 yr MI, CVA, death) (2) versus 25 age/gender-matched controls HO Learning Cohort 50 cases (3 yr MI, CVA, death) (1) versus 50 age/gender-matched controls O a 4 Time (h) Figure 2 | Identification of metabolites of dietary PC and an obligatory role for gut flora in generation of plasma analytes associated with CVD risks. a, Summary schematic indicating structure of metabolites and routes (oral or i.p.) of formation observed in choline challenge studies in mice using the indicated isotope-labelled choline. The m/z in plasma observed for the isotopomers of the choline metabolites are shown. b, Plasma levels of d9 metabolites after i.p. challenge with d9(trimethyl)dipalmitoylphosphatidylcholine (d9-DPPC). c, d9-TMAO production after oral d9-DPPC administration in mice, following suppression of gut flora with antibiotics (3 weeks), and then following placement (4 weeks) into conventional cages with non-sterile mice (‘conventionalized’). Data are presented as mean 6 standard error (s.e.) from four independent replicates. 5 8 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1 ©2011 Macmillan Publishers Limited. All rights reserved ARTICLE RESEARCH Dietary PC metabolites predict CVD risk We next sought to independently confirm the prognostic value of monitoring fasting plasma levels of TMAO, choline and betaine in a large independent clinical cohort distinct from subjects examined in both the Learning and Validation Cohorts. Stable subjects (N 5 1,876) undergoing elective cardiac evaluations were enrolled. Clinical, demographic and laboratory characteristics of the cohort are provided in Supplementary Table 4a. Fasting plasma levels of TMAO, choline and betaine were quantified by LC/MS/MS using methods specific for each analyte (Supplementary Fig. 8). Increasing levels of choline, TMAO and betaine were all observed to show dose-dependent associations with the presence of CVD (Fig. 3a–c) and multiple individual CVD phenotypes including peripheral artery disease (PAD), coronary artery disease (CAD), and history of myocardial infarction (see Supplementary Table 5a–d for multilogistic regression models, and Supplementary Table 5e for Somers’ Dxy correlation). The associations between increased risk of all CVD phenotypes monitored and elevated systemic levels of the three PC metabolites held true after adjustments for traditional cardiac risk factors and medication usage (Fig. 3a–c and Supplementary Table 5a–e). Dietary choline or TMAO promotes atherosclerosis We next investigated whether the strong associations noted between plasma levels of the dietary PC metabolites and CVD risk reflected some hidden underlying pro-atherosclerotic mechanism. Atherosclerosisprone mice (C57BL/6J Apoe2/2) at time of weaning were placed on either normal chow diet (contains 0.08–0.09% total choline, wt/wt) or normal chow diet supplemented with intermediate (0.5%) or high amounts of additional choline (1.0%) or TMAO (0.12%). At 20 weeks of age increased total aortic root atherosclerotic plaque area was noted in both male and female mice on diets supplemented with either choline or TMAO (Fig. 3d and Supplementary Fig. 9a). Analysis of plasma levels of choline and TMAO in each of the dietary arms showed nominal changes in plasma levels of choline, but significant increases of TMAO in mice receiving either choline or TMAO supplementation (Supplementary Fig. 10). Parallel examination of plasma cholesterol, triglycerides, lipoproteins, glucose levels and hepatic triglyceride content in the mice failed to show significant increases that could account for the enhanced atherosclerosis (Supplementary Table 6 and Supplementary Fig. 11). Interestingly, all dietary groups of mice revealed a significant positive correlation between plasma levels of TMAO and atherosclerotic plaque size (Fig. 3e and Supplementary Fig. 9b). Of note, plasma TMAO levels observed within the female mice (which CVD P < 0.001 16 b c 8 2 1 8 Odds ratio Odds ratio 4 4 2 0.5 0.4 0 15 30 0.3 Lesion (mm2) P = 0.01 120,000 80,000 40,000 f R = 0.42, P < 0.001 0.2 0.1 0 50 100 Human P < 0.0001 7 0 ) ) 10 22 = = (n (n 14 0 175 350 TMAO (μM) None Single Double Triple Coronary vessel disease 0. 12 % TM ho C AO e lin e lin 1% C ho C ho w (n (n = = 10 31 ) ) 0 5% 0 Betaine (μM) Mouse e P = 0.009 0. 5 10 20 40 80 TMAO (μM) Mouse P = 0.01 160,000 2 0.5 1 2 0 Choline (μM) d 4 1 1 0.5 0.4 CVD P < 0.001 12 16 8 Odds ratio CVD P < 0.001 32 TMAO (μM) a Lesion (μm2) used as isotope tracer for feeding studies. When mice were fed through oral gavage with d9-PC, the time-dependent appearance of the anticipated d9 isotopomer of TMAO was observed in plasma (Fig. 2c). Interestingly, pre-treatment of mice with a three-week course of broad-spectrum antibiotics to suppress intestinal flora completely suppressed the appearance of d9-TMAO in plasma after oral d9-PC administration (Fig. 2c). A similar pattern was observed after oral administration of d9-choline to mice, with d9-TMAO produced in untreated mice, but not in the same mice after a 3-week course of broad-spectrum antibiotics (Supplementary Fig. 7b), or in germ-free mice born sterilely by Caesarean section (Supplementary Fig. 7c). In a final series of studies, mice with suppressed intestinal microflora after antibiotics were placed in conventional cages with normal (nongerm-free) mice to permit intestinal colonization with microbes. After four weeks, repeat oral d9-PC challenge of the now ‘conventionalized’ mice resulted in readily detectable plasma levels of d9-TMAO (Fig. 2c). Similar results were observed after conventionalization of germ-free mice and oral d9-choline (Supplementary Fig. 7c). Collectively, these results show an obligate role for intestinal microbiota in the generation of TMAO from the dietary lipid PC. They also reveal the following metabolic pathway for dietary PC producing TMAO: PC R choline R TMA R TMAO. Figure 3 | Plasma levels of choline, TMAO and betaine are associated with atherosclerosis risks in humans and promote atherosclerosis in mice. a–c, Spline models of the logistic regression analyses reflecting risk of CVD (with 95% CI) according to plasma levels of choline, TMAO and betaine in the entire cohort (n 5 1,876 subjects). d, Comparison in aortic lesion area among 20-week-old female C57BL/6J Apoe2/2 mice fed with chow diet supplemented with the indicated amounts (wt/wt) of choline or TMAO from time of weaning (4 weeks). e, Relationship between plasma TMAO levels and aortic lesion area. f, Relationship between fasting plasma levels of TMAO versus CAD burden among subjects (N 5 1,020). Boxes represent 25th, 50th and 75th percentile, and whiskers 5th and 95th percentile plasma levels. Single, double and triple coronary vessel disease refers to number of major coronary vessels demonstrating $50% stenosis on diagnostic coronary angiography. get enhanced atherosclerosis relative to their male counterparts), even on normal chow diet, were significantly higher than those observed among male mice (Supplementary Fig. 10). No significant gender differences in plasma levels of TMAO were observed in humans (P 5 0.47); however, a clear dose–response relationship was observed between TMAO levels and clinical atherosclerotic plaque burden in subjects undergoing coronary angiography (Fig. 3f). Hepatic FMOs, TMAO and atherosclerosis Hepatic FMO3 is a known enzymatic source for TMAO in humans, based on the recent recognition of the aetiology of an uncommon genetic disorder called trimethylaminuria (also known as fish malodour syndrome)15,17. Subjects with this metabolic condition have impaired capacity to convert TMA, which smells like rotting fish, into TMAO, an odourless stable oxidation product17. We therefore sought to identify possible sources of genetic regulation and the role of Fmo3 in atherosclerosis using integrative genetics in mice18. Expression levels of Fmo3 were determined by microarray analysis in the livers of mice from an F2 intercross between atherosclerosis-prone C57BL/ 6J Apoe2/2 mice and atherosclerosis-resistant C3H/HeJ Apoe2/2 mice and compared with quantitative measures of atherosclerosis. The expression level of Fmo3 showed marked differences between genders (females .1,000 fold higher than in males). Significant positive correlations were readily found between hepatic Fmo3 expression and atherosclerotic lesions (Fig. 4a, Supplementary Fig. 12, top row, and Supplementary Fig. 13). Interestingly, a highly significant negative correlation with plasma high-density lipoprotein (HDL) cholesterol levels was noted in both male and female mice (Fig. 4b and Supplementary Fig. 12, middle row). Further, plasma levels of the PC metabolite TMAO showed a significant positive correlation with hepatic Fmo3 expression level in mice (Fig. 4c and Supplementary Fig. 12, bottom row). 7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 5 9 ©2011 Macmillan Publishers Limited. All rights reserved RESEARCH ARTICLE Female 6,000 0 –2 –0.5 1 2.5 Fmo3 expression (Mean Log ratio) –2 –1 0 1 Fmo3 expression (Mean Log ratio) e Mouse P = 0.0006 Aortic lesion (μm2) 500,000 s /H 3,000 3H er C 1,000 2,000 Liver FMO3 mRNA eJ 0 et H 0 ou 0 yg R = 0.49 P = 0.07 250,000 oz 7 P = 0.0014 J Human 14 Plasma TMAO (μM) 10 0 –2 –1 0 1 Fmo3 expression (Mean Log ratio) d 20 /6 0 R = 0.80 P < 0.001 BL 0.4 c 30 Female R = –0.44 P < 1 × 10–6 57 12,000 C b R = 0.29 P = 0.002 HDL cholesterol (mg ml–1) Lesion (mm2) 0.8 TMAO (μM) Female a Apoe–/– f Mouse peritoneal macrophages g Mouse peritoneal macrophages P < 0.001 P < 0.001 P = 0.001 P < 0.001 ) 10 (n e in ta Be 0% 1. 0. = 10 = (n TM % C 0% AO e lin w ho ho C ) ) 10 ) (n = (n (n e in ta Be = 10 10 ) ) 0 = 10 = (n AO TM 0% 1. % 12 0. 0% C C ho ho lin w e (n (n = = 10 10 ) ) 0 0.2 12 0.1 1. Surface CD36 protein 0.2 1. 0.4 Surface SR-A1 protein P = 0.008 0.3 P = 0.04 significantly correlated with aortic lesion formation, HDL cholesterol concentrations and plasma TMAO levels (Supplementary Figs 14–16), suggesting that several members of the FMO family of enzymes may participate in atherosclerosis and the PC R TMAO metabolic pathway. To explore the relationship between hepatic FMOs and plasma TMAO levels in humans, paired samples of liver and plasma from subjects undergoing elective liver biopsy were examined. Among all of the human FMO genes monitored, only a trend towards positive association was noted between hepatic expression of FMO3 and plasma TMAO levels (Fig. 4d and Supplementary Fig. 17). Next, we focused on the genetic regulation of hepatic Fmo3 expression (and other Fmo genes) using expression quantitative trait locus (eQTL) analyses in the F2 mouse intercross. The eQTL plot for Fmo3 messenger RNA levels is shown in Supplementary Fig. 18, and demonstrates a strongly suggestive cis locus (lod score 5 5.9) on mouse chromosome 1 at 151 Mb. Fmo3 (and several other Fmo genes) is located at 164.8 Mb in a region identified as non-identical by descent between C3H/HeJ and C57BL/6 (http://mouse.cs.ucla.edu/perlegen/). This region is just distal to the 95% confidence interval of a previously reported murine atherosclerosis susceptibility locus20. Examining the effect of the closest single-nucleotide polymorphism (SNP) to Fmo3 (rs3689151) as a function of alleles inherited from either parental strain indicated a strong effect on atherosclerosis in both genders of the F2 mice (Kruskal–Wallis test, P , 1.0 3 1026). Bonferroni corrected pairwise comparisons indicated a dose-dependent significant increase in atherosclerosis in F2 mice heterozygous or homozygous for the C57BL/6J allele (Fig. 4e). Although the resolution on average for an F2 intercross of this size is in excess of 20 Mb and thus does not provide ‘gene-level’ resolution, these data show that the locus encompassing the Fmo gene cluster on chromosome 1 is associated with atherosclerotic lesion size. Collectively, these results indicate that: (1) hepatic expression levels of multiple Fmo genes are linked to plasma TMAO levels in mice; (2) hepatic expression levels of multiple Fmo genes are associated with both the extent of aortic atherosclerosis and HDL cholesterol levels in mice; (3) hepatic expression levels of FMO3 indicate an association with plasma TMAO levels in humans; and (4) a genetic locus containing the Fmo gene cluster on chromosome 1 in mice has a strong effect on atherosclerosis. Figure 4 | Hepatic Fmo genes are linked to atherosclerosis and dietary PC metabolites enhance macrophage scavenger receptor expression. a–c, Correlation between hepatic Fmo3 expression and aortic lesion, plasma HDL cholesterol and TMAO in female mice from the F2 intercross between atherosclerosis-prone C57BL/6J Apoe2/2 and atherosclerosis-resistant C3H/ HeJ Apoe2/2 mice. d, Correlation between human hepatic FMO3 expression and plasma TMAO. e, Effect of Fmo3 genotype (SNP rs3689151) on aortic sinus atherosclerosis in male mice from the C57BL6/J Apoe2/2 and C3H/HeJ Apoe2/2 F2 intercross. f, g, Quantification of scavenger receptor CD36 and SRA1 surface protein levels in macrophages harvested from C57BL/6J mice (13 week) after three weeks of standard chow versus chow supplemented with the indicated amounts (wt/wt) of choline, TMAO or betaine. Data are presented as mean 6 s.e. from the indicated numbers of mice in each group. FMO3 is one member of a family of FMO enzymes, the majority of which are physically located as a cluster of genes on chromosome 1 in both humans and mice. The various FMOs share sequence homology and overlapping substrate specificities. Further, although rare mutations in or near the FMO3 gene have been identified in individuals with trimethylaminuria19, the impact of these mutations on other FMO genes remains unknown. Examination of the hepatic expression levels of the various FMO genes revealed that many are highly correlated with each other in both mice and humans (Supplementary Table 7). Examination of hepatic expression levels of additional Fmo genes in mice from the atherosclerosis F2 intercross revealed that multiple Fmo genes are Diet and gut flora alter macrophage phenotype To explore potential mechanisms through which dietary choline and its metabolites might exert their pro-atherosclerotic effects, C57BL/6J Apoe2/2 mice at time of weaning were placed on a normal chow diet supplemented with either choline, TMAO or betaine (for .3 weeks). Both mRNA levels (Supplementary Fig. 19) and surface protein levels (Fig. 4f, g and Supplementary Fig. 20) of two macrophage scavenger receptors implicated in atherosclerosis, CD36 and SR-A1, were then determined in peritoneal macrophages. Relative to normal chow diet, mice supplemented with either choline, TMAO or betaine all showed enhanced macrophage levels of CD36 and SR-A1. We next examined the impact of dietary choline and gut flora on endogenous formation of cholesterol-laden macrophage foam cells, one of the earliest cellular hallmarks of the atherosclerotic process. Hyperlipidaemic C57BL/6J Apoe2/2 mice were fed diets with defined levels of choline as follows: (1) ‘control’ (0.07–0.08%, wt/wt), which is similar to the choline content of normal chow (0.08–0.09%); versus (2) high ‘choline’, corresponding to a .10-fold higher level of choline (1.0%, wt/wt) than normal chow. Concomitantly, half of the mice were administered broad-spectrum antibiotics for 3 weeks to suppress intestinal microflora, which was confirmed by the reduction of plasma TMAO levels by .100-fold (plasma TMAO concentrations in groups receiving antibiotics were ,100 nM). Whereas mice on the control diet showed modest evidence of endogenous macrophage foam cell formation, as indicated by Oilred-O staining of peritoneal macrophages, mice on the 1% choline supplemented diet showed markedly enhanced lipid-laden macrophage development (Fig. 5a). In contrast, suppression of intestinal flora 6 0 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1 ©2011 Macmillan Publishers Limited. All rights reserved ARTICLE RESEARCH significantly inhibited dietary-choline-induced macrophage foam cell formation (Fig. 5a, b). These results were confirmed by microscopic quantification of endogenous foam cell levels (Fig. 5b) and analytical quantification of the cholesterol content of recovered macrophages (Fig. 5c). Histopathology and biochemical studies of livers recovered from these mice showed no evidence of steatosis (Supplementary Fig. 21). Parallel analyses of plasma PC metabolites also demonstrated no significant changes in choline or betaine levels between the different dietary groups, and significant increases of plasma TMAO levels only in mice on the high-choline diet in the absence of antibiotics (males, –Abx 6 P = 0.02 2 500 μm 500 μm 0 –Abx P = 0.01 P = 0.04 8 4 P = 0.09 0 Control Choline +Abx P = 0.37 15 P = 0.003 P = 0.02 10 5 P = 0.47 0 Control Choline (n 12) = 14 ) (n = 11 (n ) = 10 ) = (n 13) = 13 ) Choline (n = (n = 24 ) 21 ) Control 12 Choline = 0 M+F (n 30,000 h P = 0.56 Aorta CD36 (×104 μm2) P = 0.33 +Abx 6) 8) Control (n –Abx = (n 13) = 17 ) 60,000 (n M+F (n 90,000 Aortic lesion (μm2) g P = 0.002 P = 0.04 (n = (n 8) = 7) P = 0.53 Female –Abx +Abx Aorta macrophage (F4/80 positive, ×104 μm2) f = P = 0.06 30,000 500 μm (n (n P < 0.0001 P = 0.01 = (n 20) = 18 ) 500 μm P = 0.13 –Abx +Abx 60,000 = (n 13) = 11 ) Choline/Abx = = 6) = e Male Choline Aortic lesion (μm2) Female mice Control Control/Abx Choline 5) Control 50 μm (n d (n P < 0.05 (n Total cholesterol (μg μg–1 DNA) 4 0 50 μm 6) 5) 6) = = –Abx +Abx P = 0.04 P < 0.001 c Apoe–/– Abx Choline diet Abx Control diet (n (n 50 μm Apoe–/– Choline 8) Control = 0 50 μm +Abx P = 0.004 P < 0.001 = Foam cells (%) 12 (n b Apoe–/– Choline diet Apoe–/– Control diet (n a Figure 5 | Obligatory role of gut flora in dietary choline enhanced atherosclerosis. a, Choline supplementation promotes macrophage foam cell formation in a gut-flora-dependent fashion. C57BL/6J Apoe2/2 mice at time of weaning (4 weeks) were provided drinking water with or without broadspectrum antibiotics (Abx), and placed on chemically defined diets similar in composition to normal chow (control diet, 0.08 6 0.01% total choline, wt/wt) or normal chow with high choline (choline diet, 1.00% 6 0.01% total choline, wt/wt). Resident peritoneal macrophages were recovered at 20 weeks of age. Typical images of Oil-red-O/haematoxylin-stained macrophages in each diet group are shown. b, Foam cell quantification from peritoneal macrophages recovered from mice in studies described in panel a. c, Macrophage cellular cholesterol content. d, Representative Oil-red-O/haematoxylin-stained aortic root sections from female C57BL/6J Apoe2/2 mice fed control and highcholine diets in the presence or absence of antibiotics. e, f, Aortic lesion area in 20 week old C57BL/6J Apoe2/2 mice off or on antibiotics and fed with control or choline diet. g, Aortic macrophage quantification with anti-F4/80 antibody staining. h, Quantification of the scavenger receptor CD36 in aorta within the indicated groups. Error bars represent s.e.m. from the indicated numbers of mice. control versus choline diet, 2.5 6 0.1 mM versus 28.3 6 2.4 mM, P , 0.001; for females, control versus choline diet, 4.0 6 0.5 mM versus 158.6 6 32.9 mM, P , 0.001). Gut flora promote diet-induced atherosclerosis In additional studies we sought to test whether gut flora is involved in dietary choline-induced atherosclerosis. At the time of weaning (4 weeks old), atherosclerosis-prone C57BL/6J Apoe2/2 mice were placed on either a control diet (0.08 6 0.01%, wt/wt, choline) or a diet supplemented with 1% choline (wt/wt, choline diet). Half of the mice were also treated with broad-spectrum antibiotics to suppress intestinal microflora. Serial plasma measurements confirmed suppression of TMAO levels to virtually non-detectable levels (,100 nM) throughout the duration of the study. At 20 weeks of age, mice were killed and aortic root lesion development was quantified. In the absence of antibiotics (that is, with preserved intestinal microflora), choline supplementation augmented atherosclerosis in both male and female mice nearly threefold (Figs 5d–f). In contrast, suppression of intestinal flora completely inhibited dietary choline-mediated enhancement in atherosclerosis (Figs 5d–f). Aortic macrophage content and scavenger receptor CD36 immunoreactive surface area within aortic lesions were markedly increased in mice on the high-choline diet, but not when intestinal microflora was suppressed with antibiotic treatment (Fig. 5g, h and Supplementary Figs 22, 23). Both histopathological and biochemical examination of liver sections from mice showed no evidence of steatosis or altered neutral lipid (triglyceride or cholesterol/cholesterol ester) levels on either diet in the absence or presence of antibiotics (Supplementary Fig. 21 and Supplementary Table 8). Finally, the structural specificity of PC metabolites in promoting a pro-atherogenic macrophage phenotype was examined. Mice fed diets supplemented with trimethylamine species (choline or TMAO) showed increased peritoneal macrophage cholesterol content and raised plasma levels of TMAO. In contrast, dietary supplementation with the choline analogue 3,3dimethyl-1-butanol (DMB), where the quaternary amine nitrogen of choline is replaced with a carbon, resulted in no TMAO increase and no increased cholesterol in macrophages (Supplementary Fig. 24). Discussion Using a targeted metabolomics approach aimed at identifying plasma metabolites the levels of which predict risk of CVD in subjects, we have identified a novel pathway linking dietary lipid intake, intestinal microflora and atherosclerosis (Fig. 6). The pathway identified (dietary PC/ choline R gut-flora-formed TMA R hepatic-FMO-formed TMAO) represents a unique additional nutritional contribution to the pathogenesis of CVD that involves PC and choline metabolism, an obligate role for the intestinal microbial community, and regulation of surface expression levels of macrophage scavenger receptors known to participate in the atherosclerotic process. The pro-atherogenic gutflora-generated metabolite TMAO is formed in a two-step process initiated by gut-flora-dependent cleavage of a trimethylamine species (for example, PC, choline, betaine) generating the precursor TMA, and subsequent oxidation by FMO3 and possibly other FMOs (Fig. 6). PC is by far the most abundant dietary source of choline in most humans. The present results indicate that both environmental exposure (dietary lipid from predominantly animal products) and microbial flora participate in TMAO formation and producing a pro-atherogenic macrophage phenotype. Although the present genetic studies also indicate a role for hepatic expression levels of one or more Fmo genes in both enhanced atherosclerotic plaque and decreased HDL levels in mice, the participation of FMO genes in human atherosclerosis and HDL cholesterol levels remains to be established. Strong associations between systemic TMAO levels and both angiographic measures of coronary artery atherosclerotic burden and cardiac risks were observed among subjects; however, no correlation was observed between plasma TMAO levels and HDL cholesterol levels in subjects. It remains to be determined whether genetic impairment in FMO3 alone or in 7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 6 1 ©2011 Macmillan Publishers Limited. All rights reserved RESEARCH ARTICLE Ph os Heart attack ne choli dyl ati ph Stroke TMA Death Gut flora Atherosclerosis FMOs TMAO Choline Fatty acid Fatty acid O O P O O- Choline CH3 + HO CH2 CH2 N CH3 CH3 Phosphatidylcholine (dietary) Gut flora Choline CH3 + H N CH3 CH3 Hepatic FMOs Trimethyl amine (TMA) CH3 + HO N CH3 CH3 Trimethylamine N-oxide (TMAO) Figure 6 | Gut-flora-dependent metabolism of dietary PC and atherosclerosis. Schematic summary illustrating newly discovered pathway for gut-floramediated generation of pro-atherosclerotic metabolite from dietary PC. combination with other FMO genes is protective for CVD. No phenotype other than the objectionable odour accompanying this disorder is known. In fact, individuals with trimethylaminuria often become vegans, as reducing the ingestion of dietary animal products rich in lipids decreases TMA production and the associated noxious odour. Little is also known about the biological functions of TMAO in humans. TMAO apparently serves as an osmolite in the freeze-avoidance response of some species21. In vitro it can function as a small-molecule chaperone, affecting the folding and functioning of some proteins22,23. In addition, TMAO and TMA accumulate in plasma of subjects on maintenance haemodialysis24, raising the possibility that TMAO may contribute to the well-established enhanced CVD risk noted among subjects with end-stage renal disease. Choline is an essential nutrient that is usually grouped within the vitamin B complex. Choline and its metabolite betaine are methyl donors, along with folate, and are metabolically linked to transmethylation pathways including synthesis of the CVD risk factor homocysteine. Deficiency in both choline and betaine have been suggested to produce epigenetic changes in genes linked to atherosclerosis25,26, and acute choline and methionine deficiency in rodent models causes lipid accumulation in liver (steatohepatitis), heart and arterial tissues27. Alternatively, some studies have reported an association between increased whole blood levels of total choline and cardiovascular disease28,29. Few clinical studies have examined the relationship between choline intake and CVD30, probably because accurate measures of the choline content of most foods has only recently become available14 (http://www.nal. usda.gov/fnic/foodcomp/Data/Choline/Choln02.pdf). The association between dietary choline (and alternative trimethyl-amine-containing species) and atherosclerosis will be complex because, as the present studies show, it will be influenced by inter-individual differences in the composition of the intestinal microflora. The human intestinal microbial community is an enormous and diverse ecosystem with known functions in nutrition, gut epithelial cell health, and innate immunity31. Intestinal flora recently also has been implicated in the development of some metabolic phenotypes such as obesity and insulin resistance, as well as alterations in immune responses11,32–34. To our knowledge, the present studies are the first to identify a direct link between intestinal microflora, dietary PC and CVD risk. These results indicate that an appropriately designed probiotic intervention may serve as a therapeutic strategy for CVD. Interestingly, production of TMAO can be altered by probiotic administration35. Thus, in addition to the current clinical recommendation for a general reduction in dietary lipids, manipulation of commensal microbial composition may be a novel therapeutic approach for the prevention and treatment of atherosclerotic heart disease and its complications. The present studies also suggest a further novel treatment for atherosclerosis—blocking the presumed pathogenic biochemical pathway at the level of the gut flora through use of a non-systemically absorbed inhibitor. METHODS SUMMARY Plasma samples and associated clinical study data were identified in patients referred for cardiac evaluation at a tertiary care centre. All subjects gave written informed consent and the Institutional Review Board of the Cleveland Clinic approved all study protocols. Unbiased metabolic profiling was performed using LC/MS. Target analyte structural identification was achieved using a combination of LC/MS/MS, LC/MSn, multinuclear NMR, GC/MS and choline isotope tracer feeding studies in mice as outlined in Methods. Statistical analyses were performed using R (version 2.10.1)36. Intestinal microflora were suppressed by supplementation of drinking water with a cocktail of broad-spectrum antibiotics37. Germ-free mice were purchased from Taconic SWGF. QTL analyses to identify atherosclerosis-related genes were performed on F2 mice generated by crossing atherosclerosis-prone C57BL/6J Apoe2/2 mice and atherosclerosis-resistant C3H/HeJ Apoe2/2 mice38. mRNA expression was assayed by microarray analysis and real-time PCR. Aortic root lesion area in mice was quantified by microscopy after staining39. Mouse peritoneal macrophages were collected by lavage for foam cell quantification and cholesterol accumulation assay. Surface protein levels of scavenger receptors CD36 and SR-A1 were determined by flow cytometry. Full Methods and any associated references are available in the online version of the paper at www.nature.com/nature. Received 29 July 2009; accepted 9 February 2011. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. Epstein, S. E. et al. The role of infection in restenosis and atherosclerosis: focus on cytomegalovirus. Lancet 348 (suppl. 1), S13–S17 (1996). Patel, P. et al. Association of Helicobacter pylori and Chlamydia pneumoniae infections with coronary heart disease and cardiovascular risk factors. Br. Med. J. 311, 711–714 (1995). Danesh, J., Collins, R. & Peto, R. Chronic infections and coronary heart disease: is there a link? Lancet 350, 430–436 (1997). Saikku, P. et al. Serological evidence of an association of a novel Chlamydia, TWAR, with chronic coronary heart disease and acute myocardial infarction. Lancet 332, 983–986 (1988). O’Connor, C. M. et al. Azithromycin for the secondary prevention of coronary heart disease events—the WIZARD study: a randomized controlled trial. J. Am. Med. Assoc. 290, 1459–1466 (2003). Cannon, C. P. et al. Antibiotic treatment of Chlamydia pneumoniae after acute coronary syndrome. N. Engl. J. Med. 352, 1646–1654 (2005). Andraws, R., Berger, J. S. & Brown, D. L. Effects of antibiotic therapy on outcomes of patients with coronary artery disease: a meta-analysis of randomized controlled trials. J. Am. Med. Assoc. 293, 2641–2647 (2005). Wright, S. D. et al. Infectious agents are not necessary for murine atherogenesis. J. Exp. Med. 191, 1437–1442 (2000). Backhed, F., Ley, R. E., Sonnenburg, J. L., Peterson, D. A. & Gordon, J. I. Hostbacterial mutualism in the human intestine. Science 307, 1915–1920 (2005). Turnbaugh, P. J. et al. An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444, 1027–1031 (2006). Dumas, M. E. et al. Metabolic profiling reveals a contribution of gut microbiota to fatty liver phenotype in insulin-resistant mice. Proc. Natl Acad. Sci. USA 103, 12511–12516 (2006). Cashman, J. R. et al. Biochemical and clinical aspects of the human flavincontaining monooxygenase form 3 (FMO3) related to trimethylaminuria. Curr. Drug Metab. 4, 151–170 (2003). Al-Waiz, M., Mikov, M., Mitchell, S. C. & Smith, R. L. The exogenous origin of trimethylamine in the mouse. Metabolism 41, 135–136 (1992). Zeisel, S. H., Mar, M. H., Howe, J. C. & Holden, J. M. Concentrations of cholinecontaining compounds and betaine in common foods. J. Nutr. 133, 1302–1307 (2003). 6 2 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1 ©2011 Macmillan Publishers Limited. All rights reserved ARTICLE RESEARCH 15. Lang, D. H. et al. Isoform specificity of trimethylamine N-oxygenation by human flavin-containing monooxygenase (FMO) and P450 enzymes: selective catalysis by fmo3. Biochem. Pharmacol. 56, 1005–1012 (1998). 16. Zhang, A. Q., Mitchell, S. C. & Smith, R. L. Dietary precursors of trimethylamine in man: a pilot study. Food Chem. Toxicol. 37, 515–520 (1999). 17. Mitchell, S. C. & Smith, R. L. Trimethylaminuria: the fish malodor syndrome. Drug Metab. Dispos. 29, 517–521 (2001). 18. Schadt, E. E. et al. An integrative genomics approach to infer causal associations between gene expression and disease. Nature Genet. 37, 710–717 (2005). 19. Dolphin, C. T., Janmohamed, A., Smith, R. L., Shephard, E. A. & Phillips, I. R. Missense mutation in flavin-containing mono-oxygenase 3 gene, FMO3, underlies fish-odour syndrome. Nature Genet. 17, 491–494 (1997). 20. Wang, S. S. et al. Identification of pathways for atherosclerosis in mice: integration of quantitative trait locus analysis and global gene expression data. Circ. Res. 101, e11–e30 (2007). 21. Treberg, J. R., Wilson, C. E., Richards, R. C., Ewart, K. V. & Driedzic, W. R. The freezeavoidance response of smelt Osmerus mordax: initiation and subsequent suppression of glycerol, trimethylamine oxide and urea accumulation. J. Exp. Biol. 205, 1419–1427 (2002). 22. Devlin, G. L., Parfrey, H., Tew, D. J., Lomas, D. A. & Bottomley, S. P. Prevention of polymerization of M and Z a1-Antitrypsin (a1-AT) with trimethylamine N-oxide. Implications for the treatment of a1-AT deficiency. Am. J. Respir. Cell Mol. Biol. 24, 727–732 (2001). 23. Song, J. L. & Chuang, D. T. Natural osmolyte trimethylamine N-oxide corrects assembly defects of mutant branched-chain a-ketoacid decarboxylase in maple syrup urine disease. J. Biol. Chem. 276, 40241–40246 (2001). 24. Bain, M. A., Faull, R., Fornasini, G., Milne, R. W. & Evans, A. M. Accumulation of trimethylamine and trimethylamine-N-oxide in end-stage renal disease patients undergoing haemodialysis. Nephrol. Dial. Transplant. 21, 1300–1304 (2006). 25. Dong, C., Yoon, W. & Goldschmidt-Clermont, P. J. DNA methylation and atherosclerosis. J. Nutr. 132, 2406S–2409S (2002). 26. Zaina, S., Lindholm, M. W. & Lund, G. Nutrition and aberrant DNA methylation patterns in atherosclerosis: more than just hyperhomocysteinemia? J. Nutr. 135, 5–8 (2005). 27. Salmon, W. D. & Newberne, P. M. Cardiovascular disease in choline-deficient rats. Effects of choline deficiency, nature and level of dietary lipids and proteins, and duration of feeding on plasma and liver lipid values and cardiovascular lesions. Arch. Pathol. 73, 190–209 (1962). 28. Danne, O., Lueders, C., Storm, C., Frei, U. & Mockel, M. Whole blood choline and plasma choline in acute coronary syndromes: prognostic and pathophysiological implications. Clin. Chim. Acta 383, 103–109 (2007). 29. LeLeiko, R. M. et al. Usefulness of elevations in serum choline and free F2isoprostane to predict 30-day cardiovascular outcomes in patients with acute coronary syndrome. Am. J. Cardiol. 104, 638–643 (2009). 30. Bidulescu, A., Chambless, L. E., Siega-Riz, A. M., Zeisel, S. H. & Heiss, G. Usual choline and betaine dietary intake and incident coronary heart disease: the Atherosclerosis Risk in Communities (ARIC) study. BMC Cardiovasc. Disord. 7, 20 (2007). 31. Eckburg, P. B. et al. Diversity of the human intestinal microbial flora. Science 308, 1635–1638 (2005). 32. Ley, R. E., Turnbaugh, P. J., Klein, S. & Gordon, J. I. Microbial ecology: human gut microbes associated with obesity. Nature 444, 1022–1023 (2006). 33. Li, M. et al. Symbiotic gut microbes modulate human metabolic phenotypes. Proc. Natl Acad. Sci. USA 105, 2117–2122 (2008). 34. Reigstad, C. S., Lunden, G. O., Felin, J. & Backhed, F. Regulation of serum amyloid A3 (SAA3) in mouse colonic epithelium and adipose tissue by the intestinal microbiota. PLoS ONE 4, e5842 (2009). 35. Martin, F. P. et al. Probiotic modulation of symbiotic gut microbial–host metabolic interactions in a humanized microbiome mouse model. Mol. Syst. Biol. 4, 157 (2008). 36. Rizzo, M. L. Statistical Computing with R (Chapman & Hall/CRC, 2008). 37. Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S. & Medzhitov, R. Recognition of commensal microflora by toll-like receptors is required for intestinal homeostasis. Cell 118, 229–241 (2004). 38. Wang, S. et al. Genetic and genomic analysis of a fat mass trait with complex inheritance reveals marked sex specificity. PLoS Genet. 2, e15 (2006). 39. Baglione, J. & Smith, J. D. Quantitative assay for mouse atherosclerosis in the aortic root. Methods Mol. Med. 129, 83–95 (2006). Supplementary Information is linked to the online version of the paper at www.nature.com/nature. Acknowledgements We wish to thank E. Sehayek for discussions, L. W. Castellani for help with lipoprotein profile analysis, and F. McNally, M. Berk and M. Pepoy for technical assistance. This research was supported by National Institutes of Health grants R01 HL103866, P01 HL098055, P01HL087018-020001, P01 HL28481 and P01 HL30568. B.J.B. was supported by NIH training grant T32-DK07789. The clinical study GeneBank was supported in part by P01 HL076491-055328, R01 HL103931 and the Cleveland Clinic Foundation General Clinical Research Center of the Cleveland Clinic/ Case Western Reserve University CTSA (1UL1RR024989). Some of the laboratory studies (haemaglobin A1C, fasting glucose) in GeneBank were supported by R01 DK080732 and Abbott Diagnostics provided supplies for performance of some of the fasting lipid profile, glucose, creatinine, troponin I and hsCRP measured in GeneBank. Author Contributions Z.W. performed metabolomics analyses, and biochemical, cellular, animal model and mass spectrometry studies. He assisted with statistical analyses, and assisted in both drafting and critical review of the manuscript. E.K., B.D. and J.D.S. assisted with performance of animal models and their analyses. B.S.L. synthesized d9-DPPC and assisted in metabolomics/mass spectrometry analyses. B.J.B., H.A. and A.J.L. performed the mouse eQTL experiments and analyses, and assisted in both drafting and critical review of the manuscript. A.J.L. provided some funding for the study. R.K., E.B.B., X.F. and Y.-M.C. performed mass spectrometry analyses of clinical samples. Y.W. performed statistical analysis. A.E.F. and P.S. helped with collection of human liver biopsy material and interpretation of biochemical and pathological examination of animal liver for steatosis. W.H.W.T. assisted in GeneBank study design and enrolment, as well as analyses of clinical studies and critical review of the manuscript. J.A.D. assisted in clinical laboratory testing for human clinical studies, animal model experimental design, and critical review of the manuscript. S.L.H. conceived of the idea, designed experiments, assisted in data analyses, the drafting and critical review of the manuscript, and provided funding for the study. Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Readers are welcome to comment on the online version of this article at www.nature.com/nature. Correspondence and requests for materials should be addressed to S.L.H. (hazens@ccf.org). 7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 6 3 ©2011 Macmillan Publishers Limited. All rights reserved RESEARCH ARTICLE METHODS 40 General procedures. Lipids were extracted by chloroform:methanol (2:1, v/v) . Cholesterol was quantified by GC/MS41. Triglyceride was quantified by GPO reagent set (Pointe Scientific)42. Cell DNA content was quantified by PicoGreen43. RNA was isolated by TRIZOL reagent (Invitrogen) and RNeasy Mini Kit (Qiagen). All reagents were purchased from Sigma unless otherwise specified. Research subjects. Plasma samples and associated clinical data were collected as part of two studies involving stable non-symptomatic subjects undergoing elective cardiac evaluations at a tertiary care centre. The first study, GeneBank, is a large well-characterized tissue repository with longitudinal data from subjects undergoing elective diagnostic left heart catheterization or elective coronary computed tomography angiography44. The second study, BioBank, includes subjects undergoing cardiac risk factor evaluation/modification in a preventive cardiology clinic45. CAD included adjudicated diagnoses of stable or unstable angina, myocardial infarction or angiographic evidence of $50% stenosis of one or more epicardial vessels. PAD was defined as any evidence of extra-coronary atherosclerosis. Atherosclerotic CVD was defined as the presence of either CAD or PAD. All subjects gave written informed consent and the Institutional Review Board of the Cleveland Clinic approved all study protocols. Discovery metabolomics analyses began with an unbiased search for plasma (fasting, EDTA purple top tube) analytes linked to CVD risk using a case–control design (Learning Cohort, N 5 50 cases and 50 controls). Cases were randomly selected from GeneBank subjects who experienced a myocardial infarction, stroke or death over the ensuing 3-year period. An age- and gender-matched control group was randomly selected from GeneBank subjects that did not experience a CVD event. An independent non-overlapping Validation Cohort (N 5 25 cases and 25 controls) was also from GeneBank. A third large (N 5 1,876) independent study comprised of non-overlapping subjects then evaluated clinical associations of identified analytes. Approximately half (N 5 1,020) of the subjects enrolled were from GeneBank and the remaining (N 5 856) were from BioBank. Similar patient characteristics within each cohort and the combined cohort are observed, as shown in Supplementary Table 4a, b. The association of each PC metabolite and various cardiovascular phenotypes within each cohort (GeneBank and BioBank) are also similar (Supplementary Tables 4c–e). All subjects in the large independent clinical study had similar inclusion and exclusion criterion, negative cardiac enzymes (troponin I , 0.03 ng ml21) and no recent history of myocardial infarction or coronary artery bypass graft. Estimate of glomerular filtration rate was calculated using the MDRD formula46. Fasting blood glucose, C reactive protein, troponin I and lipid profiles were measured on the Abbott ARCHITECT platform (Abbott Diagnostics). Metabolomics analyses. Plasma proteins were precipitated with four volumes of ice-cold methanol and small-molecule analytes within supernatants were analysed after injection onto a phenyl column (4.6 3 250 mm, 5 mm Rexchrom Phenyl; Regis) at a flow rate of 0.8 ml min21 using a Cohesive HPLC interfaced with a PE Sciex API 365 triple quadrupole mass spectrometer (Applied Biosystems) with Ionics HSID1, EP101, XT1 redesigned source and collision cell as upgrades in positive MS1 mode. LC gradient (LC1) starting from 10 mM ammonium formate over 0.5 min, then to 5 mM ammonium formate, 25% methanol and 0.1% formic acid over 3 min, held for 8 min, followed by 100% methanol and water washing for 3 min at a flow rate of 0.8 ml min21 was used to resolve analytes. Spectra were continuously acquired after the initial 4 min. Peaks within reconstructed ion chromatograms at 1 AMU increments were integrated and both retention times and m/z of analytes were used for statistical analyses. Selection criteria for determining analytes of interest were based on the composite of MACE as the clinical phenotype, defined as incident myocardial infarction, stroke or death at 3 years, and included: (1) demonstration of a statistically significant difference between cases versus controls using a Bonferroni adjusted two sided t-test (P , 0.05); (2) evidence of a significant (P , 0.05) dose–response relationship between analyte level and clinical phenotype using Cochran– Armitage trend test; and (3) a minimal signal-to-noise ratio of 5:1 for a given analyte. Chemical characterization of unknown metabolites. To chemically define the structures of the plasma analytes selected for further investigation (that is, analytes with m/z 76, 104 and 118 in positive MS1 mode), multiple approaches were used. Analytes of interest were isolated by HPLC, vacuum dried, re-dissolved in water and injected onto the same phenyl column with a distinct HPLC gradient (LC2, flow rate: 0.8 ml min21) starting from 0.2% formic acid over 2 min, then linearly to 18% acetonitrile containing 0.2% formic acid over 18 min and further to 100% acetonitrile containing 0.2% formic acid over 3 min. The targeted analytes were identified by their m/z and the appropriate fractions recovered. After removal of solvent, dry analytes were used for structural identification. Samples analysed by GC/MS were derivatized using Sylon HTP kit (HMDS 1 TMCS 1 Pyridine (3: 1: 9), Supelco). Derivatization of TMAO and the plasma analyte at m/z 76 also included initial reduction by titanium (III) chloride47 and further reaction with 2,2,2-trichloroethylchloroformate48. Analyses were performed on the Agilent Technolgies 6890/5973 GC/MS in positive ion chemical ionization mode. The GC/MS analyses used a J&W Scientific DB-1 column (30 m, 0.25-mm inner diameter, 0.25-mm film thickness) for separations. Quantification of TMAO, choline and betaine. Stable isotope dilution LC/MS/ MS was used for quantification of TMAO, choline and betaine. TMAO, choline and betaine were monitored in positive MRM MS mode using characteristic precursor–product ion transitions: m/z 76 R 58, m/z 104 R 60 and m/z 118 R 59, respectively. The internal standards TMAO-trimethyl-d9 (d9TMAO) and choline-trimethyl-d9 (d9-choline), were added to plasma samples before protein precipitation, and were similarly monitored in MRM mode at m/z 85 R 68 and m/z 113 R 69, respectively. Various concentrations of TMAO, choline and betaine standards and a fixed amount of internal standards were spiked into control plasma to prepare the calibration curves for quantification of plasma analytes. TMA was similarly quantified from acidified plasma by LC/MS/MS using MRM mode. Aortic root lesion quantification. Apolipoprotein E knockout mice (C57BL/6J Apoe2/2) were weaned at 4 weeks of age and fed with either standard chow control diet (Teklad 2018) or a custom diet comprised of normal chow supplemented with 0.5% choline (Teklad TD.07863), 1.0% choline (Teklad TD.07864) or 0.12% TMAO (Teklad TD.07865) for 16 weeks. Mice were anaesthetized with ketamine/xylazine before cardiac puncture to collect blood. Hearts were fixed and stored in 4% paraformaldehyde before frozen OCT sectioning and staining with Oil red O and haematoxylin. Aortic root lesion area was quantified as the mean value of six sections39. Aortic sections were immunostained with rat anti-mouse F4/80 antibody (ab6640, Abcam) followed by goat anti-rat IgG-FITC antibody (sc-2011, Santa Cruz) and FITC-conjugated CD36 monoclonal antibody (Cayman Chemical) for F4/80 and CD36, respectively. Sections were mounted in Vectashield DAPI (H-1200, Vectashield) to take pictures under a Leica DMR microscope (W. Nuhsbaum) equipped with a Q Imaging Retiga EX camera. We used Image-Pro Plus Version 7.0 (MediaCybernetics) to integrate the positive staining area of F4/80 and CD36 in aorta. Flow cytometry assays on scavenger receptors. Cell surface expression of scavenger receptors SR-A1 and CD36 were quantified on peritoneal macrophages from female mice by flow cytometry after immunostaining with fluorochromeconjugated antibodies. Fluorescence intensity was quantified on a FACSCalibur flow cytometry instrument with FlowJo software (BD Biosciences). More than 10,000 total events were acquired to obtain adequate macrophages numbers. The following antibodies were used to stain macrophages: CD36 monoclonal antibody FITC (Cayman Chemical), anti-mouse SR-AI/MSRA1 (R&D Systems), goat anti-rat IgG-FITC (Santa Cruz Biotechnology), Alexa Fluor 647 anti-mouse F4/80 (eBioscience), Alexa Fluor 647 anti-mouse CD11b (eBioscience) and the isotype controls, Alexa Fluor 647 rat IgG2b (eBioscience), Alexa Fluor 647 rat IgG2a (eBioscience), normal mouse IgA-FITC (Santa Cruz). Cells were incubated with antibodies for 30 min at 4uC and washed with 0.1% BSA in PBS. Cells with double-positive staining for F4/80 and CD11b were gated as macrophage49–51 for the quantification of fluorescence intensity for CD36 and SR-A1 (Supplementary Fig. 20), with results normalized to F4/80. eQTL studies. C57BL/6J Apoe2/2 (B6 Apoe2/2) mice were purchased from the Jackson Laboratory and C3H/HeJ Apoe2/2 (C3H Apoe2/2) mice were bred by backcrossing B6 Apoe2/2 to C3H/HeJ for 10 generations. F2 mice were generated by crossing B6 Apoe2/2 with C3H Apoe2/2 and subsequently intercrossing the F1 mice. Mice were fed Purina Chow containing 4% fat until 8 weeks of age, and then transferred to a Western diet (Teklad 88137) containing 42% fat and 0.15% cholesterol for 16 weeks until euthanasia at 24 weeks of age. Mouse atherosclerotic lesion area was quantified using standard methods39. eQTL analyses were performed as previously described38. Each individual sample was hybridized against the pool of F2 samples. Significantly differentially expressed genes were determined as previously described52. Expression data in the form of mean log ratios (mlratios) were treated as a quantitative trait in eQTL analysis using Rqtl package for the R language and environment for statistical computing (http://cran.r-project.org/). Germ-free mice and conventionalization studies. An antibiotic cocktail (0.5 g l21 vancomycin, 1 g l21 neomycin sulphate, 1 g l21 metronidazole, 1 g l21 ampicillin) previously shown to be sufficient to deplete all detectable commensal bacteria37 was administered in drinking water ad libitum. In additional studies, 8-week-old female Swiss Webster germ-free mice (Taconic SWGF) underwent an oral (gavage) choline challenge (see later) immediately after their removal from their germ-free microisolator shipper. After the choline or PC challenge, the germ-free mice were placed in conventional cages with non-sterile C57BL/6J female mice to facilitate transfer of commensal organisms. Four weeks later, the conventionalized mice underwent a second choline or PC challenge. ©2011 Macmillan Publishers Limited. All rights reserved ARTICLE RESEARCH In vivo macrophage studies. C57BL/6J mice or B6 Apoe 2/2 mice were fed with either standard chow control diet (Teklad 2018) or a custom diet supplemented with 1.0% betaine (Teklad TD.08112), 1.0% choline (Teklad TD.07864), 0.12% TMAO (Teklad TD.07865) or 1.0% dimethylbutanol (DMB) supplemented in drinking water for at least 3 weeks. Elicited mouse peritoneal macrophages (MPMs) were harvested by peritoneal lavage with ice-cold PBS 3 days after i.p. injection of 1 ml 4% thioglycollate. Some studies with mice were performed using a custom diet with low but sufficient choline content (0.07% total; Teklad TD.09040) versus high-choline diet (1.0% total; Teklad TD.09041) in the presence or absence of antibiotics. Choline content of all diets was confirmed by LC/MS/MS. Foam cell staining. Foam cells were identified by microscopy cultured peritoneal macrophages on glass coverslips after 6 h in RPMI 1640 medium supplemented with 3% lipoprotein-deficient serum. Cells were fixed with paraformaldehyde and stained with Oil red O/haematoxylin53. Cells containing .10 lipid droplets were scored as foam cells50. At least 10 fields and 500 cells per condition were counted. Real-time PCR. Real-time PCR of Cd36, Sr-a1 and flavin monooxygenases (mouse Fmos) was performed using Brilliant II SYBR Green qRT–PCR kit (Strategene). The forward and reverse primers Cd36, Gapdh, Sr-a1, mouse Fmos and F4/80 were synthesized by IDT based on sequences reported54–58. RT–PCR of human FMOs was similarly performed using primers specific for the sequence of each of the indicated human FMOs. d9-DPPC synthesis and vesicle preparation. d9-DPPC was synthesized by reacting 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine (Genzyme Pharmaceuticals) withper-deuteromethyliodide(CD3I,CambridgeIsotopeLaboratories)59,60.Theproduct waspurifiedbypreparativesilicagelTLCandconfirmedbybothMSandNMR.Eggyolk lecithin (Avanti Polar Lipids) and d9-DPPC liposomes used for gavage feeding and i.p. injection of mice were prepared by the method of extrusion through polycarbonate filters61. Metabolic challenges in mice. C57BL/6J mice were administered (gavage) unlabelled or the indicated stable-isotope-labelled choline or PC (egg yolk lecithin or d9-DPPC) using a 1.5-inch 20-gauge intubation needle. Choline challenge: gavage consisted of 150 ml of 150 mM d9-choline. PC challenge: gavage or i.p. injection of 300 ml 5 mg ml21 of unlabelled PC or labelled d9-DPPC. Mice were fasted 12 h before PC challenge. Plasma (50 ml) was collected via the saphenous vein from mice at baseline and after gavage or i.p. injection time points. Statistical analysis. Student’s t-test and Wilcoxon rank sum test were employed to compare group means62,63. Pearson correlation, Spearman rank correlation and Somers’ Dxy correlation were used to investigate the correlation between two variables64,65. Comparison of categorical measures between independent groups was done using x2 tests66. Odds ratios and 95% confidence intervals for cardiovascular phenotypes (history of myocardial infarction, CAD, PAD, CVD and CAD1PAD) were calculated with R, version 2.10.1 (http://www.r-project.org), using logistic regression67 with case status as the dependent variable and plasma analyte as independent variable. Trend tests in frequencies across quartiles were done using Cochran–Armitage trend tests68. Levels of analytes were adjusted for traditional CAD risk factors in a multivariate logistic regression model including individual traditional cardiac risk factors (age, gender, diabetes, smoking, hypertension, lipids, CRP and estimated creatinine clearance) and medication usage (statin or other lipid-lowering agents, antihypertensive agents including angiotensin-converting-enzyme inhibitor, angiotensin-receptor blocking agent, diuretic, calcium-channel blocker or beta blocker, and aspirin or other platelet inhibitors). 40. Folch, J., Lees, M. & Sloane Stanley, G. H. A simple method for the isolation and purification of total lipides from animal tissues. J. Biol. Chem. 226, 497–509 (1957). 41. Robinet, P., Wang, Z., Hazen, S. L. & Smith, J. D. A simple and sensitive enzymatic method for cholesterol quantification in macrophages and foam cells. J. Lipid Res. 51, 3364–3369 (2010). 42. Millward, C. A. et al. Genetic factors for resistance to diet-induced obesity and associated metabolic traits on mouse chromosome 17. Mamm. Genome 20, 71–82 (2009). 43. Ahn, S. J., Costa, J. & Emanuel, J. R. PicoGreen quantitation of DNA: effective evaluation of samples pre- or post-PCR. Nucleic Acids Res. 24, 2623–2625 (1996). 44. Wang, Z. et al. Protein carbamylation links inflammation, smoking, uremia and atherogenesis. Nature Med. 13, 1176–1184 (2007). 45. Nicholls, S. J. et al. Lipoprotein (a) levels and long-term cardiovascular risk in the contemporary era of statin therapy. J. Lipid Res. 51, 3055–3061 (2010). 46. Stoves, J., Lindley, E. J., Barnfield, M. C., Burniston, M. T. & Newstead, C. G. MDRD equation estimates of glomerular filtration rate in potential living kidney donors and renal transplant recipients with impaired graft function. Nephrol. Dial. Transplant. 17, 2036–2037 (2002). 47. Barham, A. H. et al. Appropriateness of cholesterol management in primary care by sex and level of cardiovascular risk. Prev. Cardiol. 12, 95–101 (2009). 48. daCosta, K. A., Vrbanac, J. J. & Zeisel, S. H. The measurement of dimethylamine, trimethylamine, and trimethylamine N-oxide using capillary gas chromatographymass spectrometry. Anal. Biochem. 187, 234–239 (1990). 49. Schledzewski, K. et al. Lymphatic endothelium-specific hyaluronan receptor LYVE1 is expressed by stabilin-11, F4/801, CD11b1 macrophages in malignant tumours and wound healing tissue in vivo and in bone marrow cultures in vitro: implications for the assessment of lymphangiogenesis. J. Pathol. 209, 67–77 (2006). 50. Cailhier, J. F. et al. Conditional macrophage ablation demonstrates that resident macrophages initiate acute peritoneal inflammation. J. Immunol. 174, 2336–2342 (2005). 51. Kunjathoor, V. V. et al. Scavenger receptors class A-I/II and CD36 are the principal receptors responsible for the uptake of modified low density lipoprotein leading to lipid loading in macrophages. J. Biol. Chem. 277, 49982–49988 (2002). 52. Yang, X. et al. Tissue-specific expression and regulation of sexually dimorphic genes in mice. Genome Res. 16, 995–1004 (2006). 53. Zhou, J., Lhotak, S., Hilditch, B. A. & Austin, R. C. Activation of the unfolded protein response occurs at all stages of atherosclerotic lesion development in apolipoprotein E-deficient mice. Circulation 111, 1814–1821 (2005). 54. Miles, E. A., Wallace, F. A. & Calder, P. C. Dietary fish oil reduces intercellular adhesion molecule 1 and scavenger receptor expression on murine macrophages. Atherosclerosis 152, 43–50 (2000). 55. Westendorf, T., Graessler, J. & Kopprasch, S. Hypochlorite-oxidized low-density lipoprotein upregulates CD36 and PPARc mRNA expression and modulates SR-BI gene expression in murine macrophages. Mol. Cell. Biochem. 277, 143–152 (2005). 56. Rasooly, R., Kelley, D. S., Greg, J. & Mackey, B. E. Dietary trans 10, cis 12-conjugated linoleic acid reduces the expression of fatty acid oxidation and drug detoxification enzymes in mouse liver. Br. J. Nutr. 97, 58–66 (2007). 57. Zhang, J. & Cashman, J. R. Quantitative analysis of FMO gene mRNA levels in human tissues. Drug Metab. Dispos. 34, 19–26 (2006). 58. de Vries, T. J., Schoenmaker, T., Hooibrink, B., Leenen, P. J. & Everts, V. Myeloid blasts are the mouse bone marrow cells prone to differentiate into osteoclasts. J. Leukoc. Biol. 85, 919–927 (2009). 59. Chen, F. C. M. & Benoiton, L. N. A new method of quatenizing amines and its use in amino acid and peptide chemistry. Can. J. Chem. 54, 3310–3311 (1976). 60. Morano, C., Zhang, X. & Fricker, L. D. Multiple isotopic labels for quantitative mass spectrometry. Anal. Chem. 80, 9298–9309 (2008). 61. Greenberg, M. E. et al. The lipid whisker model of the structure of oxidized cell membranes. J. Biol. Chem. 283, 2385–2396 (2008). 62. Gauvreau, K. & Pagano, M. Student’s t-test. Nutrition 9, 386 (1993). 63. Wijnand, H. P. & van de Velde, R. Mann–Whitney/Wilcoxon’s nonparametric cumulative probability distribution. Comput. Methods Programs Biomed. 63, 21–28 (2000). 64. Gaddis, M. L. & Gaddis, G. M. Introduction to biostatistics: part 6, correlation and regression. Ann. Emerg. Med. 19, 1462–1468 (1990). 65. Deichmann, M. et al. S100-b, melanoma-inhibiting activity, and lactate dehydrogenase discriminate progressive from nonprogressive American Joint Committee on Cancer stage IV melanoma. J. Clin. Oncol. 17, 1891–1896 (1999). 66. Goodall, C. M., Stephens, O. B. & Moore, C. M. Comparative sensitivity of survivaladjusted chi-square and normal statistics for the mutagenesis fluctuation assay. J. Appl. Toxicol. 6, 95–100 (1986). 67. Traissac, P., Martin-Prevel, Y., Delpeuch, F. & Maire, B. Logistic regression vs other generalized linear models to estimate prevalence rate ratios. Rev. Epidemiol. Sante Publique 47, 593–604 (1999). 68. Gautam, S. Test for linear trend in 2 3 K ordered tables with open-ended categories. Biometrics 53, 1163–1169 (1997). ©2011 Macmillan Publishers Limited. All rights reserved SUPPLEMENTARY INFORMATION doi:10.1038/nature09922 Supplementary Table 1. Characteristics of the 18 analytes monitored by LC-MS with unique m/z and specific retention time in positive ion mode which meet acceptability criteria in both the Learning Cohort andValidation Cohort. Analyte (m/z) 76 87 101 104 105 115 116 118 130 133 138 139 147 147 176 178 191 196 Retention time (min) 7.5 8.6 4.3 9.3 9.3 5.8 4.8 6.2 4.3 4.4 6.5 6.5 4.3 5.6 5.8 5.8 5.5 6.8 Control 1.6E+08 6.2E+07 1.6E+07 5.6E+08 1.4E+08 4.4E+07 3.3E+07 9.7E+07 8.4E+07 5.5E+06 4.3E+07 6.4E+06 2.3E+08 1.4E+07 1.4E+07 4.3E+06 4.7E+06 7.7E+06 Peak area Cases Adjusted -Log(P) 2.4E+08 1.85 7.8E+07 1.64 1.9E+07 3.59 6.6E+08 3.00 1.9E+08 3.71 6.0E+07 2.52 4.5E+07 4.52 4.1E+08 3.90 9.5E+07 3.28 6.9E+06 2.40 8.3E+07 3.13 1.5E+07 3.29 2.6E+08 3.91 1.8E+07 3.86 2.0E+07 2.52 1.5E+07 1.54 7.1E+06 4.30 1.3E+08 1.74 Odds ratio (95% CI) P for trend Hazard ratio (95% CI) 7.7 (2.2-27.4) 8.1 (2.2-29.5) 9.5 (2.6-34.9) 8.1 (2.2-29.5) 8.1 (2.2-29.5) 4.5 (1.3-15.3) 9.5 (2.6-34.9) 2.4 (1.1-5.3) 7.7 (2.2-27.4) 8.1 (2.2-29.5) 5.1 (1.5-17.4) 8.1 (2.2-29.5) 8.1 (2.2-29.5) 5.1 (1.5-17.4) 27.9 (5.9-132.3) 8.0 (2.2-29.2) 8.0 (2.2-29.3) 5.3 (1.5-18.2) <0.01 <0.01 <0.01 <0.01 <0.01 0.01 <0.01 0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 0.01 8.4 (2.5-27.8)** 6.8 (2.2-20.5)** 7.4 (2.2-24.7)** 18.0 (4.9-66.5)** 20.8 (5.6-77.1)** 5.1 (1.6-16.1)** 13.0 (3.8-44.6)** 3.9 (1.3-12.0)** 8.8 (2.6-29.6)** 5.2 (1.8-15.2)** 7.5 (2.2-25.6)** 4.2 (1.7-10.5)** 9.7 (2.9-32.7)** 6.4 (2.0-20.6)** 19.3 (5.7-65.2)** 4.7 (1.6-14.1)** 1.3 (0.4-3.8) 4.8 (1.5-15.0)* Plasma samples for the metabolomics discovery studies were all from the clinical study Gene Bank, a large and well-characterized tissue repository with longitudinal data from sequential consenting subjects undergoing elective diagnostic left heart catheterization. The initial Learning Cohort was comprised of 50 cases (randomly selected from amongst GeneBank subjects who suffered from non-fatal MI, stroke or death within 3 years after enrollment) and 50 age-gender matched controls that did not experience non-fatal MI, stroke or death in the 3 years following enrollment. The peaks in LC chromatograms were integrated and compared between controls and cases using a Bonferoni adjusted two sided t-test and adjusted p values (for multiple sampling) were calculated. Adjusted logP values >1.3 we reconsidered significant. Odds ratios (OR) and 95% confidence intervals (CI) comparing top quartile versus lowest quartile to predict risk for non-fatal MI, stroke or death in 3 years were calculated using logistic regression and odds ratios with 95% CI that failed to cross unity were consider significant. Trend tests in frequencies across quartiles were done using Cochran-Armitage trend tests, and p<0.05 was consider significant. Only analytes with signal to noise ratio of 5 or greater were included in analyses. A second Validation Cohort (n=25 cases and 25 controls) from GeneBank that is non-overlapping, and used similar selection criterion for detecting analytes associated with CVD risks was also examined. A total of 18 analytes met all selection criteria within both the initial Learning Cohort, and the subsequent ValidationCohort. Hazard ratios and 95% CI for analytes as continuous variables were also calculated using Cox regression. Data in the table are shown for cases vs. controls from the combined Learning and Validation cohorts, including the m/z, retention time, peak areas, OR(95% CI), and Hazard ratio (95% CI). *p<0.05; **p<0.01. WWW.NATURE.COM/ NATURE | 1 Supplementary Table 2. GC-MS analysis of candidate compounds with m/z=76 in positive ion mode. System 1, trimethylchlorosilane (TMCS); System 2, thrichloroethyl chloroformate (TCECF) Component System1 Derivative RT Product (min) ions System2 Derivative RT (min) Product ions Component Purified from Plasma ND N, N-dimthyl trichloroethylcarbmate 4.3 219, 184, 131, 117 95, 88 72, 61, 44 TMAO ND N, N-dimthyl trichloroethylcarbmate 4.3 219, 184, 131, 117 95, 88 72, 61, 44 1-Amino-2propanol TMS- 4.4 147, 130, 116, 102 86, 73, 59, 44 ND 2-Amino-1propanol DiTMS- 3.8 220, 204, 188, 147 132, 116 93, 73, 59, 44 ND 3-Amino-1propanol DiTMS- 3.7 220, 204, 188, 147 132, 116 93, 73, 59, 44 ND Methylaminoethanol TMS- 5.2 147, 131, 117, 100, 73, 59, 44 ND Glycolamide DiTMS- 5.3 220, 204, , 188, 132, 73, 59, 44 ND Hydroxyguanidine TMS- 4.3 147, 131, 117, 100, 87, 73, 59, 45 ND Glycine TriTMS- 5.3 294, 278, 220, 188, 147, 131, 116, 73, 59, 45 ND 4.3 147, 131, 117, 100, 73, 59, 45 ND N-IsopropylTMShydroxylamine WWW.NATURE.COM/ NATURE | 2 Supplementary Table 3. GC-MS analysis of candidate compounds with m/z=104 in positive ion mode. Candidate compounds were analyzed following derivitization with trimethylsilyl using Sylon HTP kit. Compound Derivative RT Product ions (min) Component from plasma Choline 2-Amino-3-Methyl-1butanol 3-Amino-isobutyric Acid DiTMS 2.6 248, 218, 188, 160, 144, 73 DiTMS 2.9 2-Aminobutyric acid DiTMS 3.0 TetraTMS 3.2 DiTMS 2.7 248, 218, 201, 174, 147, 130, 102, 73, 45 248, 220, 174, 147, 131, 100, 73, 45 391, 345, 329, 242, 215, 189, 172, 147, 131, 100, 73, 45 248, 218, 190, 174, 147, 131, 100, 73, 45 Diethylenetriamine TetraTMS 3.4 392, 354, 282, 258, 231, 208, 189, 171, 143, 126, 102, 73, 56 Ethyl-N-hydroxylacetimidate benzonitrile TMS 2.6 177, 160, 144, 125, 117, 100, 73, 45 DiTMS 2.7 2-Isopropylaminoethanol TMS 5.6 247, 223, 188, 171, 147, 130, 79, 73, 42 175, 103, 72, 44 1-dimethylaminopropanol TMS 2.6 Biuret N,N-Dimethylglycine 2-propylaminoethanol _ 176, 160, 144, 126, 117, 100, 73, 45 WWW.NATURE.COM/ NATURE | 3 Supplementary Table 4a . Subject characteristics for entire cohort examined Characteristic Whole cohort (n=1876) Age 64.2 ± 10.3 Women (%) 51.2 Diabetes (%) 31.2 Hypertension (%) 70.1 History of smoking (%) 59.4 Current smoking (%) 10 LDL 98 (81- 120) HDL 35 (29- 43) Triglycerides 115 (83- 164) CRP 2.4 (1.0- 5.9) Framingham Risk Score 10.0 (7.0- 12.0) MDRD (GFR) 73.0 (60.4- 86.0) ACE (%) 47.6 STATIN (%) 52.8 ASPIRIN (%) 69.0 TMANO 3.9 (2.6- 6.3) Choline 10.7 (8.7- 13.4) Betaine 41.2 (33.1- 51.0) CVD (%) 65.3 PAD (%) 23.1 CAD (%) 42.2 History of MI (%) 27.0 Values expressed in mean ± standard deviation or median (interquartile range). WWW.NATURE.COM/ NATURE | 4 Supplementary Table 4b. Comparison of clinical characteristics (Genebank vs. BioBank subjects) Characteristic GeneBank BioBank P value (n=1,020) (n=856) 63.9 ± 9.3 64.5±11.5 0.221 Women (%) 50.7 51.7 0.681 Diabetes (%) 33.9 28.0 0.07 Hypertension (%) 68.5 72.0 0.116 History of smoking (%) 55.0 64.6 <0.001 LDL 98 (81- 121) 99 (80- 120) 0.996 HDL 35 (28- 43) 36 (30- 43) 0.06 Triglycerides 117 (83- 166) 112 (82- 163) 0.597 CRP 2.5 (1.0- 5.8) 2.3 (1.0- 5.9) 0.743 Framingham Risk Score 10.0 (7.0- 12.0) 10.0 (7.0- 12.0) 0.690 MDRD (GFR) 72.6 (60.6- 85.5) 73.8 (60.2- 86.5) 0.358 ACE inhibitors (%) 48.2 46.9 0.583 Statins (%) 51.1 54.8 0.121 Aspirin (%) 67.7 70.5 0.204 CVD (%) 65.4 65.2 0.941 PAD (%) 24.5 21.5 0.140 CAD (%) 41.0 43.7 0.249 Age Plasma from subjects examined in the third large independent cohort for verification of the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank and BioBank have similar inclusion criterion. Each is comprised of stable consenting subjects undergoing elective cardiac evaluation for either cardiovascular risk factor evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank). The clinical characteristics of subjects analyzed from GeneBank and BioBank are listed above, and reveal similar overall clinical characteristics for cardiac risk factors and prevalent cardiovascular disease. Values expressed in mean ± standard deviation or median (interquartile range). WWW.NATURE.COM/ NATURE | 5 Supplementary Table 4c. Comparison of odds ratio and 95% confidence intervals (CIs) for association between plasma choline levels and the indicated cardiovascular phenotypes within Genebank and BioBank subjects GeneBank (n=1,020) Characteristic BioBank (n=856) Unadjusted Adjusted Unadjusted Adjusted 1.36 (0.95-1.95) (p=0.09) 1.76 (1.23-2.54) (p=0.002) 2.98 (2.03-4.38) (p<0.001) 1.22 (0.82-1.81) (p=0.33) 1.36 (0.91-2.04) (p=0.13) 1.91 (1.24-2.96) (p=0.004) 1.79 (1.21-2.64) (p=0.004) 1.80 (1.21-2.66) (p=0.003) 3.23 (2.13-4.91) (p<0.001) 1.67 (1.10-2.54) (p=0.02) 1.56 (1.03-2.38) (p=0.04) 2.70 (1.71-4.27) (p<0.001) 1.43 (0.88-2.32) (p=0.15) 2.51 (1.57-4.01) (p<0.001) 4.92 (3.06-7.91) (p<0.001) 1.38 (0.79-2.42) (p=0.25) 1.88 (1.08-3.27) (p=0.03) 2.87 (1.61-5.12) (p<0.001) 1.44 (0.81-2.53) (p=0.21) 1.99 (1.14-3.47) (p=0.02) 4.31 (2.48-7.48) (p<0.001) 1.20 (0.65-2.20) (p=0.56) 1.60 (0.88-2.91) (p=0.12) 2.91 (1.57-5.38) (p=0.001) 1.45 (0.86-2.46) (p=0.17) 2.80 (1.70-4.62) (p<0.001) 5.69 (3.44-9.39) (p<0.001) 1.43 (0.76-2.71) (p=0.27) 2.00 (1.06-3.76) (p=0.03) 3.07 (1.60-5.89) (p=0.001) 1.30 (0.72-2.37) (p=0.38) 1.97 (1.10-3.50) (p=0.02) 4.01 (2.27-7.08) (p<0.001) 1.04 (0.55-1.99) (p=0.89) 1.45 (0.78-2.72) (p=0.24) 2.47 (1.30-4.68) (p=0.006) 1.36 (0.96-1.94) (p=0.09) 1.74 (1.22-2.50) (p=0.003) 2.92 (1.99-4.28) (p<0.001) 1.24 (0.84-1.83) (p=0.27) 1.36 (0.92-2.02) (p=0.12) 1.93 (1.26-2.97) (p=0.003) 1.86 (1.26-2.76) (p=0.002) 1.86 (1.26-2.76) (p=0.002) 3.39 (2.23-5.15) (p<0.001) 1.76 (1.16-2.67) (p=0.008) 1.64 (1.08-2.49) (p=0.02) 2.94 (1.86-4.62) (p<0.001) 1.57 (0.98-2.51) (p=0.06) 1.68 (1.05-2.69) (p=0.03) 2.64 (1.67-4.17) (p<0.001) 1.29 (0.78-2.13) (p=0.31) 1.16 (0.70-1.92) (p=0.57) 1.52 (0.92-2.5) (p=0.10) 1.60 (1.05-2.43) (p=0.03) 1.32 (0.86-2.02) (p=0.20) 2.13 (1.40-3.25) (p<0.001) 1.60 (1.02-2.49) (p=0.04) 1.23 (0.78-1.93) (p=0.37) 2.02 (1.27-3.22) (p=0.003) CAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 PAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 CAD + PAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 CVD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 MI Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 Plasma from subjects examined in the third large independent cohort for verification of the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank and BioBank have similar inclusion criterion. Each is comprised of stable consenting subjects undergoing elective cardiac evaluation for either cardiovascular risk factor evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank). Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma choline levels within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal function and medication use, as indicated in Supplemental Methods. Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 6 Supplementary Table 4d. Comparison of odds ratio and 95% confidence intervals (CIs) for association between plasma TMAO levels and the indicated cardiovascular phenotypes within Genebank and BioBank subjects GeneBank (n=1,020) Characteristic Unadjusted Adjusted BioBank (n=856) Unadjusted Adjusted CAD Quartile 2 vs 1 1.33 (0.93-1.90) (p=0.12) 1.56 (1.18-2.07) (p=0.002) 1.86 (1.26-2.75) (p=0.002) 1.85 (1.22-2.8) (p=0.004) Quartile 3 vs 1 1.59 (1.11-2.29) (p=0.01) 3.53 (2.36-5.26) (p<0.001) 1.44 (1.08-1.91) (p=0.01) 2.59 (1.88-3.55) (P<0.001) 1.87 (1.27-2.77) (p=0.002) 3.74 (2.44-5.72) (p<0.001) 1.71 (1.12-2.61) (p=0.01) 3.08 (1.94-4.88) (p<0.001) 1.15 (0.70-189) (p=0.58) 2.11 (1.32-3.36) Quartile 3 vs 1 (p=0.002) 6.01 (3.72-9.72) Quartile 4 vs 1 (p<0.001) CAD + PAD 1.49 (0.99-2.24) (p=0.05) 1.66(1.11-2.49) (p=0.01) 3.43(2.26-5.21) (p<0.001) 2.17 (1.23-3.84) (p=0.008) 1.97 (1.11-3.48) (p=0.02) 4.96 (2.82-8.73) (p<0.001) 2.07( 1.12-3.82) (p=0.02) 1.66 (0.89-3.09) (p=0.11) 3.75 (2.00-7.03) (p<0.001) Quartile 2 vs 1 1.36 (0.79-2.34) (p=0.26) 2.58 (1.55-4.28) (p<0.001) 7.47 (4.45-12.52) (p<0.001) 1.86 (1.18-2.93) (p=0.008) 1.85 (1.17-2.91) (p=0.009) 4.03 (2.54-6.40) (p<0.001) 2.51 (1.37-4.59) (p=0.003) 2.12 (1.15-3.92) (p=0.02) 5.60 (3.08-10.2) (p<0.001) 2.44 (1.26-4.71) (p=0.008) 1.79 (0.91-3.52) (p=0.09) 4.00 (2.04-7.82) (p<0.001) 1.26 (0.89-1.80) (p=0.50) 1.53 (1.07-2.18) (p=0.02) 3.41 (2.29-5.06) (P<0.001) 1.50 (1.14-1.98) (p=0.004) 1.39 (1.05-1.83) (p=0.02) 2.54 (1.86-3.47) (p<0.001) 1.80 (1.22-2.66) (p=0.003) 1.87 (1.26-2.75) (p=0.002) 3.70 (2.42-5.64) (p<0.001) 1.78 (1.18-2.69) (p=0.006) 1.70 (1.12-2.58) (p=0.01) 3.08 (1.96-4.86) (p<0.001) 0.98 (0.60-1.59) (p=0.93) 1.86 (1.19-2.90) (p=0.006) 2.23 (1.43-3.46) (p<0.001) 0.95 (0.57-1.59) (p=0.86) 1.53 (0.95-2.46) (p=0.08) 1.47 (0.90-2.40) (p=0.12) 1.48 (0.97-2.27) (p=0.07) 1.37 (0.89-2.1) (p=0.15) 2.47 (1.62-3.75) (p<0.001) 1.51 (0.96-2.36) (p=0.07) 1.30 (0.82-2.05) (p=0.26) 2.11 (1.33-3.34) (p=0.001) Quartile 4 vs 1 PAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 CVD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 MI Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 Plasma from subjects examined in the third large independent cohort for verification of the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank and BioBank have similar inclusion criterion. Each is comprised of stable consenting subjects undergoing elective cardiac evaluation for either cardiovascular risk factor evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank). Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma TMAO levels within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal function and medication use, as indicated in Supplemental Methods. Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 7 Supplementary Table 4e. Comparison of odds ratio and 95% confidence intervals(CIs) for association between plasma betaine levels and the indicated cardiovascular phenotypes within Genebank and BioBank subjects. Characteristic GeneBank (n=1,020) BioBank (n=856) Unadjusted Adjusted Unadjusted Adjusted 1.48 (1.03-2.14) (p=0.04) 1.41 (0.98-2.03) (p=0.07) 1.63 (1.13-2.35) (p=0.009) 1.62 (1.20-2.17) (p=0.001) 1.36 (1.01-1.82) (p=0.04) 2.30 (1.69-3.14) (p<0.001) 1.88 (1.26-2.79) (p=0.002) 1.29 (0.87-1.9) (p=0.20) 3.23 (2.11-4.94) (p<0.001) 1.88 (1.22-2.90) (p=0.004) 1.30 (0.84-1.99) (p=0.24) 3.66 (2.28-5.86) (p<0.001) 1.50 (0.95-2.38) (p=0.09) 1.78 (1.14-2.79) (p=0.01) 1.87 (1.18-2.94) (p=0.007) 1.40 (0.93-2.09) (p=0.10) 1.57 (1.04-2.36) (p=0.03) 1.87 (1.24-2.83) (p=0.003) 1.48 (0.86-2.55) (p=0.16) 1.65 (0.96-2.83) (p=0.07) 2.40 (1.41-4.08) (p=0.001) 1.50 (0.82-2.74) (p=0.19) 1.67 (0.91-3.07) (p=0.10) 2.75 (1.49-5.08) (p=0.001) 1.92 (1.16-3.17) (p=0.01) 2.23 (1.37-3.64) (p=0.001) 2.53 (1.55-4.14) (p<0.001) 1.71 (1.10-2.67) (p=0.02) 1.76 (1.12-2.76) (p=0.01) 2.14 (1.37-3.36) (p=0.001) 1.34 (0.76-2.35) (p=0.32) 1.64 (0.94-2.86) (p=0.08) 2.27 (1.31-3.93) (p=0.003) 1.36 (0.72-2.58) (p=0.34) 1.64 (0.86-3.13) (p=0.13) 2.51 (1.31-4.79) (p=0.005) 1.38 (0.96-1.98) (p=0.08) 1.34 (0.93-1.92) (p=0.11) 1.49 (1.04-2.14) (p=0.03) 1.57 (1.18-2.10) (p=0.001) 1.38 (1.03-1.84) (p=0.03) 2.21 (1.63-3.00) (p<0.001) 1.84 (1.24-2.72) (p=0.003) 1.25 (0.85-1.83) (p=0.25) 3.35 (2.18-5.13) (p<0.001) 1.83 (1.20-2.80) (p=0.005) 1.28 (0.84-1.95) (p=0.25) 3.84 (2.40-6.15) (p<0.001) 1.42 (0.90-2.23) (p=0.13) 1.29 (0.81-2.04) (p=0.29) 1.72 (1.09-2.69) (p=0.02) 1.02 (0.62-1.67) (p=0.95) 0.86 (0.52-1.43) (p=0.56) 1.10 (0.66-1.82) (p=0.71) 1.30 (0.86-1.96) (p=0.22) 1.22(0.80-1.86) (p=0.35) 1.95 (1.29-2.95) (p=0.002) 1.33 (0.85-2.07) (p=0.22) 1.22 (0.77-1.92) (p=0.40) 2.09 (1.33-3.28) (p=0.001) CAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 PAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 CAD + PAD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 CVD Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 MI Quartile 2 vs 1 Quartile 3 vs 1 Quartile 4 vs 1 Plasma from subjects examined in the third large independent cohort for verification of the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank and BioBank have similar inclusion criterion. Each is comprised of stable consenting subjects undergoing elective cardiac evaluation for either cardiovascular risk factor evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank). Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma betaine levels within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal function and medication use, as indicated in Supplemental Methods. Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 8 Supplementary Table 5a. Odds ratio (95% CI) of cardiovascular phenotypes based on quartiles of plasma levels of choline. (n=1876) Risk MI Model 1 unadjusted 1 adjusted 1 unadjusted 1 PAD adjusted 1 unadjusted 1 CAD adjusted 1 unadjusted 1 PAD+ CAD adjusted 1 unadjusted 1 CVD adjusted 1 2 1.30 (0.96-1.76) (p=0.09) 1.26 (0.31-1.71) (p=0.75) 1.82 (1.25-2.66) (p=0.002) 1.51 (0.98-2.34) (p=0.06) 1.58(1.21-2.06) (p=0.001) 1.48(1.09-2.02) (p=0.01) 1.89 (1.27-2.84) (p=0.002) 1.47 (0.90-2.40) (p=0.12) 1.68 (1.29-2.18) (p<0.001) 1.56 (1.16-2.10) (p=0.009) Quartiles 3 1.21 (0.89-1.64) (p=0.22) 1.15 (0.85-1.57) (p=0.37) 2.23 (1.54-3.24) (p<0.001) 1.52 (0.99-2.35) (p=0.06) 1.66 (1.27-2.17) (p<0.001) 1.30 (0.95-1.77) (p=0.10) 2.31 (1.55-3.44) (p<0.001) 1.49 (0.92-2.41) (p=0.10) 1.73 (1.33-2.26) (p<0.001) 1.35(1.00-1.83) (p=0.05) 4 1.50 (1.11-2.03) (p=0.008) 1.39 (1.03-1.89) (p=0.03) 5.39 (3.70-7.84) (p<0.001) 3.46 (2.21-5.41) (p<0.001) 2.83 (2.14-3.76) (p=0.001) 1.92 (1.38-2.69) (p<0.001) 5.44 (3.67-8.07) (p<0.001) 3.02 (1.84-4.95) (p<0.001) 3.02 (2.28-3.99) (p<0.001) 2.18 (1.57-3.02) (p<0.001) P for trend 0.02 <0.001 <0.001 <0.001 <0.001 Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking, diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA), β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 9 Supplementary Table 5b. Odds ratio (95% CI) of cardiovascular phenotypes based on quartiles of plasma levels of TMAO. (n=1876) Risk MI PAD CAD PAD+ CAD CVD Model unadjusted 1 1 adjusted 1 unadjusted 1 adjusted 1 unadjusted 1 adjusted 1 unadjusted 1 adjusted 1 unadjusted 1 adjusted 1 2 1.39 (1.01-1.91) (p=0.04) 1.37 (1.00-1.89) (p=0.05) 1.66 (1.13-2.43) (p=0.009) 1.58(1.02-2.44) (p=0.04) 1.57 (1.20-2.04) (p=0.001) 1.66 (1.22-2.26) (p=0.001) 2.05 (1.35-3.11) (p=0.001) 2.07 (1.25-3.43) (p=0.005) 1.52 (1.17-1.97) (p=0.002) 1.56 (1.16-2.11) (p=0.003) Quartiles 3 1.54 (1.12-2.11) (p=0.008) 1.48 (1.08-2.04) (p=0.01) 2.26 (1.55-3.28) (p<0.001) 1.78 (1.15-2.77) (p=0.01) 1.70 (1.30-2.22) (p<0.001) 1.59 (1.16-2.18) (p=0.004) 2.62 (1.74-3.96) (p<0.001) 2.10 (1.26-3.5) (p=0.004) 1.63 (1.25-2.11) (p<0.001) 1.5 (1.11-2.04) (p=0.008) 4 2.23 (1.64-3.03) (p<0.001) 2.08 (1.52-2.84) (p<0.001) 5.83 (4.00-8.49) (p<0.001) 4.32 (2.73-6.84) (p<0.001) 3.55 (2.65-4.74) (p<0.001) 3.25 (2.29-4.63) (p<0.001) 7.23 (4.81-10.9) (p<0.001) 5.71 (3.36-9.7) (p<0.001) 3.46 (2.59-4.62) (p<0.001) 3.06 (2.17-4.31) (p<0.001) P for trend <0.001 <0.001 <0.001 <0.001 <0.001 Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking, diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA), β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 10 Supplementary Table 5c. Odds ratio (95% CI) of cardiovascular phenotypes based on quartiles of plasma levels of betaine. (n=1876) Risk MI Model 1 unadjusted 1 adjusted 1 unadjusted 1 PAD adjusted 1 unadjusted 1 CAD adjusted 1 unadjusted 1 PAD+ CAD adjusted 1 unadjusted 1 CVD adjusted 1 2 1.49 (1.09-2.03) (p=0.01) 1.52 (1.11-2.08) (p=0.009) 1.48 (1.03-2.11) (p=0.03) 1.15 (0.75-1.78) (p=0.52) 1.56 (1.19-2.04) (p=0.001) 1.51(1.1-2.08) (p=0.01) 1.73 (1.18-2.53) (p=0.005) 1.43 (0.88-2.34) (p=0.15) 1.56 (1.20-2.04) (p=0.001) 1.45(1.07-1.98) (p=0.02) Quartiles 3 1.45 (1.06-1.99) (p=0.02) 1.50 (1.09-2.06) (p=0.01) 1.74 (1.22-2.47) (p=0.002) 1.38 (0.89-2.15) (p=0.15) 1.39 (1.06-1.81) (p=0.02) 1.35 (0.98-1.87) (p=0.07) 1.95 (1.34-2.85) (p=0.001) 1.60 (0.97-2.65) (p=0.07) 1.40 (1.07-1.82) (p=0.01) 1.37(1.00-1.87) (p=0.05) 4 1.88 (1.38-2.56) (p<0.001) 1.96 (1.44-2.68) (p<0.001) 2.18 (1.53-3.11) (p<0.001) 1.65(1.06-2.56) (p=0.03) 2.26 (1.71-2.99) (p<0.001) 2.16 (1.54-3.03) (p<0.001) 2.52 (1.73-3.66) (p<0.001) 2.00 (1.21-3.31) (p=0.007) 2.20 (1.67-2.9) (p<0.001) 2.00 (1.44-2.78) (p<0.001) P for trend <0.001 <0.001 <0.001 <0.001 <0.001 Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking, diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA), β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios (95% CIs) are shown in bold with p values in red. WWW.NATURE.COM/ NATURE | 11 Supplementary Table 5d. Odds ratio (95% CI) of cardiovascular phenotypes based on plasma levels of choline, TMAO and betaine (as continuous variables, n=1876) Risk Model unadjusted MI adjusted unadjusted PAD adjusted unadjusted CAD adjusted unadjusted PAD+ CAD adjusted unadjusted CVD adjusted choline 1.23(1.10-1.37) (p<0.001) 1.20 (1.07-1.33) (p=0.001) 2.01(1.74-2.32) (p<0.001) 1.68(1.43-1.97) (p<0.001) 1.62(1.45-1.80) (p<0.001) 1.43(1.28-1.61) (p<0.001) 2.05(1.77-2.39) (p<0.001) 1.64(1.38-1.94) (p<0.001) 1.62(1.46-1.80) (p<0.001) 1.45(1.3-1.63) (p<0.001) TMAO 1.37(1.23-1.52) (p<0.001) 1.34(1.20-1.49) (p<0.001) 2.11(1.82-2.45) (p<0.001) 1.71(1.46-2) (p<0.001) 1.73(1.55-1.94) (p<0.001) 1.53(1.36-1.73) (p<0.001) 2.24(1.91-2.62) (p<0.001) 1.78(1.5-2.12) (p<0.001) 1.72(1.53-1.92) (p<0.001) 1.52(1.35-1.72) (p<0.001) betaine 1.36(1.22-1.53) (p<0.001) 1.38(1.23-1.55) (p<0.001) 1.43(1.26-1.63) (p<0.001) 1.32(1.13-1.54) (p<0.001) 1.43(1.29-1.58) (p<0.001) 1.41(1.25-1.57) (p<0.001) 1.53(1.33-1.77) (p<0.001) 1.41(1.19-1.67) (p<0.001) 1.4(1.27-1.54) (p<0.001) 1.38(1.23-1.54) (p<0.001) Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking, diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA), β-blocker, and renal function (estimated creatinine clearance). WWW.NATURE.COM/ NATURE | 12 T able 5e. Somers' Dxy rank correlation between plasma analytes and cardiac phenotypes (n=1876). Cardiac Phenotype Analyte Dxy p-value MI MI MI Choline TMAO Betaine 0.084 0.184 0.139 0.006 <0.001 <0.001 CAD CAD CAD Choline TMAO Betaine 0.228 0.263 0.172 <0.001 <0.001 <0.001 PAD PAD PAD Choline TMAO Betaine 0.342 0.364 0.177 <0.001 <0.001 <0.001 CAD+PAD CAD+PAD CAD+PAD Choline TMAO Betaine 0.349 0.393 0.205 <0.001 <0.001 <0.001 CVD CVD CVD Choline TMAO Betaine 0.231 0.259 0.163 <0.001 <0.001 <0.001 Somers’ Dxy rank correlation was used to examine the association between the continuous variables (plasma levels of analytes) including choline, TMAO (trimethylamine N-oxide) and betaine, and the binary variables (cardiac phenotypes) including MI (myocardial infarction), CAD (coronary artery disease), PAD (peripheral artery disease), CVD (cardiovascular disease) and the combined presence of CAD and PAD (CAD+PAD). WWW.NATURE.COM/ NATURE | 13 Supplementary Table 6. Plasma and liver lipids and glucose in 20 week C57BL/6J.Apoe-/- mice on either normal chow or normal chow supplemented with the indicated concentrations of choline or TMAO. male mice Lipids Triglyceride (mg/dL) Cholesterol (mg/dL) Plasma Liver Chow (n=9) 114±16 426±28 LDLc (mg /dL) 324±36 HDLc (mg /dL) 61.3±7.3 Glucose (mg/dL) 238±28 Triglyceride (mg/g protein) Cholesterol (mg/g protein) 34.9±4.4 10.1±0.8 0.5% choline (n=10) 150±22 (p=0.20) 387±25 (p=0.32) 291±26 (p=0.50) 63.4±5.8 (p=0.83) 290±18 (p=0.15) 42.9±8.2 (p=0.47) 9.6±0.6 (p=0.64) 1.0% choline (n=10) 99±13 (p=0.46) 346±10 (p=0.02) 308±29 (p=0.74) 54.8±6.8 (p=0.55) 191±33 (p=0.29) 48.0±5.3 (p=0.07) 10.4±0.4 (p=0.72) 0.12% TMAO (n=13) 104±8 (p=0.58) 352±12 (p=0.04) 285±13 (p=0.38) 49.6±4.0 (p=0.23) 248±20 (p=0.78) 47.2±8.5 (p=0.22) 12.0±1.2 (p=0.18) 1.0% choline (n=11) 97±7 (p=0.57) 384±12 (p=0.46) 234±78 (p=0.70) 46.7±6.5 (p=0.91) 238±10 (p=0.93) 85.5±14.9 (p=0.08) 13.7±1.1 (p=0.40) 0.12% TMAO (n=10) 104±10 (p=0.99) 347±15 (p=0.53) 271±24 (p=0.84) 31.6±8.5 (p=0.28) 238±21 (p=0.96) 103.7±26.1 (p=0.10) 14.3±1.1 (p=0.21) female mice Lipids Triglyceride (mg/dL) Chow (n=11) 103±9 Cholesterol (mg/dL) 364±23 LDLc (mg/dL) 271±24 Plasma HDLc (mg/dL) Liver 48.2±9.6 Glucose (mg/dL) 239±16 Triglyceride (mg/g protein) Cholesterol (mg/g protein) 53.8±8.1 12.6±0.8 0.5% choline (n=10) 101±12 (p=0.88) 356±25 (p=0.81) 251±12 (p=0.50) 34.2±7.5 (p=0.31) 230±12 (p=0.63) 85.4±11.8 (p=0.06) 14.4±1.1 (p=0.20) Data are presented as mean±SE from the indicated numbers of mice in each group. P values were given by student’s t test with 2 tails versus normal chow feeding group in the same gender. WWW.NATURE.COM/ NATURE | 14 Supplementary Table 7. Correlation of hepatic gene exression levels for seven (mouse, n=285) or six (human, n=465) members of the Fmo Gene Family. Mouse Fmo1 Fmo2 Fmo3 Fmo4 Fmo5 Fmo6 Fmo9 Fmo1 1 0.76 (<0.001) 0.75 (<0.001) 0.78 (<0.001) -0.47 (<0.001) -0.35 (<0.001) 0.001 (0.98) Fmo2 Fmo3 Fmo4 Fmo5 Fmo6 0.85 (<0.001) 0.73 (<0.001) -0.64 (<0.001) -0.36 (<0.001) -0.06 (0.31) 1 0.78 (<0.001) -0.71 (<0.001) -0.26 (<0.001) -0.022 (0.72) 1 -0.57 (<0.001) -0.35 (<0.001) -0.069 (0.25) 1 0.11 (0.06) 0.069 (0.24) 1 Fmo1 1 0.03 (0.55) 0.37 (<0.001) 0.26 (<0.001) 0.25 (<0.001) 0.03 (0.59) Fmo2 Fmo3 Fmo4 Fmo5 Fmo6 Fmo9 1 0.092 (0.12) 1 Human Fmo1 Fmo2 Fmo3 Fmo4 Fmo5 Fmo6 1 -0.15 (0.001) -0.26 (<0.001) -0.10 (0.03) -0.19 (<0.001) 1 0.52 (<0.001) 0.82 (<0.001) -0.12 (0.009) 1 0.46 (<0.001) 0.25 (<0.001) 1 -0.19 (<0.001) 1 Data are shown as correlation coefficient values (r) between hepatic mRNA levels of the indicated FMO enzymes. Human hepatic Fmos expression data were taken from a publicly available dataset of Caucasian liver samples, as previously reported by Schadt et al. (PLoS Biol, 2008. 6:e107). Prior to analysis, expression data for each gene was averaged across two or more probes present on the custom Agilent microarray. WWW.NATURE.COM/ NATURE | 15 Supplementary Table 8. Liver lipids in 20 week C57BL/6J.Apoe-/- mice on either control diet (choline%, 0.08± 0.01) or choline supplemented diet (choline%, 1.0%) off and on antibiotics (Abx). Male mice Control (n=11) Choline (n=5) 36.1±8.2 1.76±0.06 diet lipids Triglyceride (mg/g liver) Cholesterol (mg/g liver) P (choline vs control) Control/Abx Choline/Abx P (choline (n=10) (n=5) /Abx vs control/Abx) 58.9±4.8 0.03 46.8±8.9 56.8±8.2 0.42 1.62±0.06 0.15 2.09±0.11 2.24±0.12 0.35 P (choline vs control) Control/Abx Choline/Abx (n=10) (n=5) Female mice diet lipids Triglyceride (mg/g liver) Cholesterol (mg/g liver) Control (n=9) Choline (n=5) P (choline /Abx vs control/Abx) 121.6±31.7 121.3±22.0 0.99 71.5±16.6 79.4±11.5 0.70 3.07±0.27 0.94 2.49±0.14 2.21±0.21 0.31 3.04±0.20 Data are presented as mean±SE from the indicated numbers of mice in each group. Liver tissues, harvested from mice, were snap-frozen and stored at -80oC until use. Cholesterol was extracted with chloroform and methanol as described in Methods and quantified by stable isotope dilution GC/MS using cholesterol-2,2,3,4,4,6-d6 as internal standard. Triglyceride was measured by GPO reagent set (Pointe Scientific) after saponification to release glycerol followed by glycerol kinase and glycerol phosphate oxidase catlayzed reaction to produce H2O2, which can react with 3-hydroxy-2,4,6-tribomobenzoic acid (TBHB) catalyzed by peroxidase to yield a red colored quinoneimine dye. The optical density at 540 nm indicates the concentration of triglyceride. WWW.NATURE.COM/ NATURE | 16 Supplementary Fig. 1 - LOG(P) a 6 5 4 3 2 1 0 Analytes presumably related via common pathway 50 100 150 200 m/z r=0.63 m/z=76 p<0.001 m/z=104 25 0. 1 r= 00 0. p< r= 0. 48 p< 0. 00 1 b m/z=118 Supplementary Figure 1. Identification of plasma analytes associated with risk for near-term heart attack, stroke and death. a, Analytes identified in m/z 50-200 range with statistically significant difference in levels among subjects who experienced a major adverse cardiovascular event (MACE = myocardial infarction (MI), stroke or death) over the ensuing 3 yr period (cases) vs. those who did not experience MACE (controls) within the Learning and Validation Cohorts. Significant differences in analyte levels between cases vs. controls were determined following adjustments for multiple sampling, with –log(p) values plotted on Y axis versus mass to charge ratio (m/z) of analyte on X axis. Filled circles represent the analytes (m/z 76, m/z 104, m/z 118), which are focused on in this study. b, A significant correlation was noted among three analytes (m/z 76, m/z 104, m/z 118) whose levels tracked with MACE risk in both Learning and Validation Cohorts, suggesting their potential involvement in a common biochemical pathway. WWW.NATURE.COM/ NATURE | 17 Supplementary Fig. 2 a MH+ Candidate plasma analyte m/z=76 Trimethylamino (CH3)3N+OH N-oxide (TMAO) CH3CH(+NH3)CH2OH 3-Amino1-propanol +NH CH CH CH OH 3 2 2 2 Methylaminoethanol CH3+NH2CH2CH2OH Glycolamide HOCH2CO+NH3 0 100 76 HONHC(+NH3)=NH Glycine +NH CH COOH 3 2 1.60 .0 .0 0 100 76 76 76 76 Hydroxyguanidine MH+=76 8.22 Relative Intensity (%) 2-Amino1-propanol LC2 MH+=76 0 100 76 CH3CH(OH)CH2+NH3 1-Amino2-propanol LC1 100 .0 0 100 . 0 100 0 100 0 100 76 0 100 76 0 100 N-Isopropyl- (CH3)2CH+NH3OH hydroxylamine 76 0 2 4 6 8 10 12 2 Retention time (min) 100 Plasma 58 59 Deriv-TMAO (standard) 100 100 Trimethylamine N-oxide (TMAO) 59 58 76 0 Relative Intensity (%) Relative Intensity (%) 76 0 - Cl- c 44 0 100 40 60 m/z 80 44 N 72 O 88 M+=219 131 117 184 d Cl Cl 40 72 88 95 M+ 219 80 M+ 131 117 120 184 6 8 10 12 Derivatized-TMAO (standard) 100 Cl Deriv-plasma m/z=76 analyte 44 0 72 88 95 95 184 131 117 O Relative Intensity (%) Analyte m/z=76 b 4 Retention time (min) 0 Derivatized-plasma analyte m/z=76 100 219 0 160 m/z 200 240 3.5 4.0 4.5 5.0 Time (min) WWW.NATURE.COM/ NATURE | 18 Supplementary Figure 2. Identification of the plasma analyte at m/z 76 associated with cardiac risk to be trimethylamine N-oxide (TMAO). a, Positive ion chromatograms and the 9 potential structures of candidate compounds with m/z 76 in positive ion mode are shown, with separations performed on a phenyl column under two distinct chromatographic conditions (LC1 and LC2, which employed methanol/ammonium formate and acetonitrile/formate gradients, respectively) developed to help distinguish these candidate species. Note that most of these analytes share the identical elemental composition and that the unidentified plasma analyte with m/z 76 shows identical retention time only to authentic TMAO standard. b, Demonstration of identical collision induced dissociation (CID) mass spectra of TMAO standard and the isolated plasma analyte with m/z 76. c, Demonstration of identical full scan mass spectra of the trichloroethyl formate derivatives of TiCl3 reduced TMAO standard and the isolated plasma analyte with m/z 76 associated with CVD risks. d, Demonstration of identical retention times of gas chromatography mass spectrometry analyses of trichloroethyl formate derivatives of TiCl3 reduced authentic TMAO standard and the plasma analyte with m/z 76 in positive ion electron impact mode. WWW.NATURE.COM/ NATURE | 19 Supplementary Figure 3 a Relative Intensity (%) 100 Relative Intensity (%) human male mouse 100 m/z=76 m/z=76 m/z=104 m/z=118 0 Relative Intensity (%) 2 6 10 Time (min) 8 12 16 Time (min) 2 6 10 Time (min) m/z=104 0 0.5 1.0 2.5 3.0 3.5 4.0 4.5 m/z=118 5.0 8.5 female mouse 100 m/z=76 7.5 8.0 8.5 7.0 7.5 8.0 8.5 2.5 3.5 4.0 4.5 5.0 5.5 6.0 6.5 2 6 10 Time (min) 8 12 16 Time (min) 2 6 10 Time (min) 7.0 m/z=118 m/z=104 0 2 6 10 Time (min) 8 12 16 Time (min) 2 6 10 Time (min) 1 0 0 1 2 3 4 Time (hr) Intensity (x105) Intensity (x105) Male mouse, 1.5 mg egg yolk PC gavage m/z=76 m/z=104 2 4 2 0 0 1 2 3 4 Time (hr) Intensity (x105) b Intensity (x105) Intensity (x105) Intensity (x105) Female mouse, 1.5 mg egg yolk PC gavage m/z=76 m/z=104 9 8 6 4 3 0 0 0 1 2 3 4 0 1 2 3 4 Time (hr) Time (hr) 16 m/z=118 8 0 0 1 2 3 4 Time (hr) m/z=118 4 2 0 0 1 2 3 4 Time (hr) WWW.NATURE.COM/ NATURE | 20 Supplementary Figure 3. Demonstration of analytes at m/z 76, 104 and 118 in plasma from mice and humans with identical retention times, and their origins from dietary phosphatidylcholine. a, Typical LC chromatograms (phenyl column) for plasma from human and both male and female C57BL/6J mice, extracted at m/z 76, 104 and 118 in positive-ion MS1 mode. b, The increased integrated peak area (intensity) in extracted ion LC chromatograms at m/z 76, 104 and 118 from mice plasma after 1.5 mg bolus supplementation (gavage) of phosphatidylcholine (egg yolk PC). Mice were fasted overnight prior to supplementation. Data are presented as mean ± SE from 3 independent replicates. WWW.NATURE.COM/ NATURE | 21 Supplementary Figure 4 a MH+ MS1, m/z=104 100 Plasma analyte m/z=104 OH N+ choline 2-amino-3-methyl1-butanol 0 100 104 OH 0 100 104 NH2 0 100 O H2N 104 OH 0 100 O NH2 O 104 NH2 N H Relative Intensity (%) O H2N biuret 104 OH 2-aminobutyric acid O N,N-dimethyl glycine N 104 OH CN 104 benzonitrile diethylenetriamine H2N ethyl-N-hydroxyl acetimidate 2-isopropylaminoethanol HO 45 N 100 40 60 c 60 80 100 choline 5x 120 104 45 60 0 40 60 80 m/z 100 6 .3 8 7 .2 2 7 .9 4 8 .8 4 9 .6 8 1 0 .9 2 1 1 .1 7 1 2 .9 1 1 3 .2 3 1 4 .1 0 1 5 .2 9 1 6 .0 8 1 7 .0 7 1 .6 4 1 .7 3 1 .9 1 2 .7 1 3 .5 7 4 .3 9 6 .1 6 9 .0 8 6 .8 8 0 .1 8 7 .6 0 1 3 .6 5 1 2 .4 6 1 4 .3 2 1 6 .2 6 1 7 .1 4 0 100 0 .2 1 2 .0 3 3 .1 6 3 .5 7 5 .6 6 5 .8 2 8 .0 9 9 .6 3 1 0 .4 6 1 1 .5 4 1 2 .3 6 1 2 .7 2 1 4 .2 7 1 5 .6 8 1 4 .8 9 1 5 .2 1 1 7 .2 2 6 .5 2 0 100 0 .3 4 7 .8 1 1 .2 6 2 .2 1 3 .0 8 3 .7 0 4 .4 4 8 .5 4 9 .3 8 1 0 .7 4 1 1 .0 4 1 3 .4 7 1 3 .8 0 1 6 .8 3 1 7 .0 4 0 100 104 0 4 8 12 16 20 Retention time (min) 4 4 .9 0 4 .4 4 5 .0 8 5 .6 4 0 .8 6 104 OH 104 5 9 .9 3 .7 9 0 .0 0 .4 5 104 120 Relative Intensity (%) Relative Intensity (%) 5x 2 .8 8 0 100 OH Analyte m/z=104 Plasma 0 100 104 N H N H 0 100 0 100 O 1-dimethyl-amino2-propanol 100 104 N-OH 2-propylaminoethanol b NH2 N H 0 100 100 d 1-dimethylamino -2-propanol 0 40 100 60 80 100 120 benzonitrile 0 40 60 80 m/z 100 120 Relative Intensity (%) 3-aminoisobutyric acid Plasma MRM, m/z=104 60 100 Plasma 0 100 choline 0 4 8 12 16 20 Retention time (min) WWW.NATURE.COM/ NATURE | 22 Supplementary Figure 4. Identification of analyte at m/z 104 to be choline. a, Candidate compounds with m/z 104 and extracted ion chromatograms in positive-ion MS1 mode of the component purified from plasma with m/z 104 associated with CVD risks, choline and other standards by reverse phase HPLC coupled to tandem mass spectrometry. Sample (20 µl) was injected onto a phenyl column (4.6 × 250 mm, 5 µm Rexchrom Phenyl) (Regis) at a flow rate of 0.8 ml/min. The separation was performed using a gradient starting from 10 mM ammonium formate over 0.5 min, then to 5 mM ammonium formate, 25 % methanol and 0.1 % formic acid over 3 min, held for 8 min, followed by 100% methanol and water washing for 3 min each. b, Demonstration of identical collision induced dissociation (CID) positive ion mass spectra of isolated human plasma analyte with m/z 104 correlated with CVD risks and choline standard. c, Demonstration of CID mass spectrum of 1-dimethylamino-2-propanol and benzonitrile, a candidate compound with similar retention time to choline, but distinct mass spectra from the human plasma analyte with m/z 104 correlated with CVD risks. d, LC/MS/MS analysis (reverse phase column) in positive-ion multiple reaction monitoring mode with parent to daughter transition, 104→60, of both the human plasma analyte with m/z 104 correlated with CVD and choline standard. WWW.NATURE.COM/ NATURE | 23 Supplementary Figure 5 CH O 3 al Or or IP 3 CH HO CH 3 3 CD CD 2 CH 3 CH 3 CD CH 3 CD2 3 2-methyl-choline-d3 m/z=121 l- se hy ra et fe M ans tr d4-choline m/z=108 1-Methyl-choline-d3 m/z=121 CH 3 CH 3 N+ CH 2 Methyltransferase CH 3 HO N+ 2 CD N+ CD CH 3 CH 3 N+ 3 betaine-d2 m/z=120 3 3 CH 2 CH [O] CH C CD HO N+ CH HO a CD CD 2 2 CH CH3 O 3 choline methyl ether-d4 m/z=122 b Plasma analyte at m/z=118 100 CID MS spectrum for plasma analyte at m/z=118 Relative Intensity (%) Relative Intensity (%) 100 0 118 58 59 0 4 6 8 10 30 60 90 Time (min) 100 120 m/z 100 CID MS spectrum for betaine at m/z=118 118 Relative Intensity (%) Relative Intensity (%) betaine at m/z=118 0 CH CH 3 3 N+ CH .N+ CH CH 3 HO CH C CH2 3 N+ CH 3 CH 3 3 2 m/z=58 O CH 3 m/z=118 m/z=59 58 59 0 4 6 8 Time (min) 10 30 60 90 120 m/z WWW.NATURE.COM/ NATURE | 24 Supplementary Figure 5. Choline isotope labeling strategy for identification of human plasma analyte with m/z 118 correlated with CVD risks as betaine. PC and choline feeding studies confirmed that the plasma analyte with m/z 118 was a metabolite of both PC and choline. To discriminate amongst multiple potential methylated choline metabolites, synthetic cholines with regiospecific isotope labeling were used since metabolites produced from the distinct choline isotopomers would generate isotopomeric forms with distinct m/z as illustrated. a, Scheme shown is for methylation of choline [1,1, 2, 2-d4 isotopomer]. Plasma was collected from mice 4 h following gavage and the m/z for each potential candidate molecule were scanned by mass spectrometry in positive MS1 mode. b, The LC chromatogram of plasma analytes extracted at m/z 118 and its CID mass spectrum show identical pattern to those of betaine. WWW.NATURE.COM/ NATURE | 25 Supplementary Fig. 6 m/z 104 (choline) 5 Fold Increase b Choline gavage 4 ) MAO T ( 6 7 3 2 1 0 118 (betaine, methylcholine, or choline methyl ether) 0 2 Time (hour) 4 Relative Ion counts a d4-Choline gavage 1.0 d2-betaine 0.5 d3-methylcholine 0 d4-choline methyl ether 0 2 4 Time (hour) Supplementary Figure 6. Metabolites of choline monitored and determination of the metabolite with m/z=118. a, Demonstration that oral (gavage) choline supplementation of mice results in time-dependent increase in plasma analytes with m/z 76, m/z 104 and m/z 118 possessing identical retention times to the analytes with m/z 76, m/z 104 and m/z 118 identified in human plasma associated with CVD risk. b, Identification of betaine as the plasma analyte with m/z 118 through use of isotope tracer studies following (4h) oral administration of choline-1,1,2,2-d4. Data are presented as mean ± SE from 4 (2 males and 2 females) independent replicates. WWW.NATURE.COM/ NATURE | 26 Supplementary Figure 7 ine a t be 0.5 TMAO TMA choline TMAO TMA 0.0 0 1 2 3 Time (h) 4 Supplementary Figure 7a. Gut flora is required for the appearance of TMA and TMAO in plasma from oral route. Plasma levels of isotope labeled choline metabolites following oral (gavage) vs. intraperitoneal (i.p.) challenge with d9(trimethyl)-choline. Note that the appearance of detectable trimethylamine (TMA) and TMAO has an obligatory requirement for oral route, whereas choline and betaine do not. For the mice receiving oral d9(trimethyl)-choline, concentrations are normalized to the isotope labeled choline at 1 hour; for the i.p. mice, concentrations of analytes are normalized to the isotope labeled betaine at 1 hour. Data presented are mean ± SE from 2 independent experiments with at least 3 mice in each experiment . Post-antibiotics 25 0 0 0 1 2 Time (h) Choline Betaine 9 TMAO 0 0 1 2 Time (h) TMAO 0 1 2 Time (h) 4 Conventionalized 18 Conc (µM) Conc (µM) 18 9 0 4 Germ - free c Choline Betaine 10 TM AO 9 18 Conc (µM) 50 Choline Betaine TM AO Conc (µM) 18 Conc (µM) Pre-antibiotics b Betaine 9 5 Choline 0 4 Conc (µM) 1.0 oral d9-choline (gavage) i.p. d9-choline chol ine e in ta be Relative Conc a 0 0 1 2 Time (h) 4 WWW.NATURE.COM/ NATURE | 27 Supplementary Figure 7b,c. Gut flora plays an obligatory role in generation of TMAO. b, Demonstration of obligatory role for gut flora in TMAO, but not choline or betaine appearance in plasma of mice following choline oral challenge. c, Independent confirmation of requirement for gut flora in TMAO production from oral choline using germ-free mice. Note that germ-free mice fail to produce TMAO following choline challenge. However, following housing of germ-free mice within conventional cages with non-sterile mice for several weeks (i.e. – “conventionalization”), there is recovery of the capacity to formTMAO from oral choline challenge. Data presented are mean ± SE from 5 independent replicates. WWW.NATURE.COM/ NATURE | 28 Supplementary Figure 8 2.5 2.0 choline-d9 Peak area of choline / Development of stable isotope dilution HPLC with on-line tandem mass spectrometry assay to quantify plasma levels of choline, TMAO and betaine 1.5 1.0 y = 0.0188x + 0.3032 R2 = 0.9995 0.5 0.0 100 2.0 1.5 1.0 y = 0.0219x + 0.0203 R2 = 0.9998 0.5 0.0 0 50 TMAO (µM) 100 1.6 TMAO-d9 Peak area of betaine / 50 choline (µM) 2.5 TMAO-d9 Peak area of TMANO / 0 1.2 0.8 y = 0.0111x + 0.3718 R2 = 0.9969 0.4 0 0 50 betaine (µM) 100 Supplementary Figure 8. Standard curves for LC/ESI/MS/MS analysis of choline, TMAO and betaine within plasma matrix. Varying levels of each analyte were spiked into 20 µl control plasma followed by precipitation with 80 µl methanol containing 0.8 nmol choline-d9 and 0.8 nmol TMAO-d9. 20 µl supernatant was loaded to LC/MS. Analyses were performed using electrospray ionization in positive-ion mode with multiple reaction monitoring of parent and characteristic daughter ions and retention times specific for components monitored. The transitions monitored were mass-to-charge ratio (m/z): m/z 104.1 → 60.1 for choline; 76.0 → 58.0 for TMAO; m/z 118.1 →59.1 for betaine; m/z 113.1 → 69.1 for choline-d9; and m/z 85.0 → 66.0 for TMAO-d9. Standard curves of choline, TMAO and betaine were generated by plotting peak area ratio versus concentration spiked into plasma for each analyte. WWW.NATURE.COM/ NATURE | 29 Supplementary Figure 9 mouse Lesion (µm2) 50000 p=0.050 p=0.045 p=0.20 mouse b 100000 R=0.68 p<0.001 Lesion (µm2) a 50000 25000 0 Chow 0.5% 1% 0.12% choline choline TMAO (n=8) (n=10) (n=10) (n=13) 0 0 20 40 TMAO (µM) Supplementary Figure 9. Dietary supplementation of mice with choline, TMAO or betaine promotes atherosclerosis and plasma TMAO is positively correlated with aortic lesion area in male mice. a, Comparison in aortic lesion area among 20 week old male C57BL/6J.Apoe- /- mice fed with chow diet supplemented with the indicated amounts (wt/wt) of choline or TMAO from time of weaning (4 weeks). b, Relationship between plasma TMAO levels and aortic lesion area. WWW.NATURE.COM/ NATURE | 30 60 Supplementary Figure 10 mouse Male choline (µM) 80 p=0.38 p=0.20 0 c 50 p=0.74 40 p=0.0001 p=0.02 p=0.005 10 0 chow 0.5% 1% 0.12% choline choline TMAO (n=9) (n=10) (n=10) (n=13) TMAO (µM) TMAO (µM) 20 25 d 30 p=0.74 p=0.0003 p=0.30 0 chow 0.5% 1% 0.12% choline choline TMAO (n=9) (n=10) (n=10) (n=13) Male Female b choline (µM) a chow 0.5% 1% 0.12% choline choline TMAO (n=11) (n=10) (n=11) (n=10) Female p=0.0002 p<0.0001 200 p<0.0001 100 0 chow 0.5% 1% 0.12% choline choline TMAO (n=11) (n=10) (n=11) (n=10) Supplementary Figure 10. Plasma levels of choline (a, b) and TMAO (c, d) in mice after supplementation with choline and TMAO. Male and female C57BL/6J.Apoe- /- mice at time of weaning (4 weeks) were placed on diets consisting of normal chow or normal chow supplemented with the indicated amount (wt/wt) of choline or TMAO. Plasma was isolated after 16 weeks on the diets at time of aorta recovery for quantification of aortic sinus lesions. Data are presented as mean ± SE for the indicated numbers of mice in each group. WWW.NATURE.COM/ NATURE | 31 Supplementary Figure 11 Male, TC Female, TC 20 25 VLDL VLDL chow C50 C100 TMAO 20 TMAO Optical Density (AU) Optical Density (AU) 16 12 LDL 8 HDL 15 LDL 10 4 5 0 0 0 10 20 30 Fraction 40 chow C50 C100 50 HDL 0 10 20 30 Fraction 40 50 Supplementary Figure 11. FPLC analysis of plasma lipoprotein on the chow and 0.5% choline (C50), 1.0% choline (C100) and 0.12% TMAO supplemented chow diets. Plasma from mice on diets consisting of normal chow (chow) or normal chow supplemented with 0.5% (wt/wt) choline (C50), 1.0% choline (C100) or 0.12% TMAO TMAO) was collected after 16 weeks on the diets at time of aorta recovery. Pooled aliquots from at least 6 mice in each group were analyzed by gel filtration FPLC for separation of lipoprotein fractions. WWW.NATURE.COM/ NATURE | 32 Supplementary Figure 12 Mouse R = 0.35 (p<1x 400000 200000 R = 0.05 (p=0.61) 250000 -2.0 -1.0 0 1.0 0 -2.0 -1.0 Fmo3 Expression (Mean Log Ratio) 1.0 -2.0 50 R = -0.19 (p=0.032) 100 0 0 -2.0 1.0 3 Plasma TMAO (µM) 0 0 1.0 -0.5 1 2.5 Fmo3 Expression (Mean Log Ratio) -1.0 0 1.0 Fmo3 Expression (Mean Log Ratio) Female 30 R = 0.85 p=0.004 2 1 0 -2 -2.0 Male R = 0.80 p<0.001 10 -1.0 Fmo3 Expression (Mean Log Ratio) Fmo3 Expression (Mean Log Ratio) Male+Female 20 R = -0.44 (p<1x 10-6) 0 Plasma TMAO (µM) -1.0 1.0 60 0 -2.0 0 Female 120 HDL Cholesterol (mg/dL) HDL Cholesterol (mg/dL) 100 -1.0 Fmo3 Expression (Mean Log Ratio) Male 200 10-15) 150 Plasma TMAO (µM) HDL Cholesterol (mg/dL) 0 Fmo3 Expression (Mean Log Ratio) Male+Female R = -0.47 (p<1x R = 0.29 (p=0.002) 400000 0 0 30 Female 800000 Aortic Lesion (µm 2) 600000 Male 500000 10-7) Aortic Lesion (µm 2) Aortic Lesion (µm 2) Male+Female R = 0.80 p<0.001 20 10 0 -2 -1.5 Fmo3 Expression (Mean Log Ratio) -1 -2 -0.5 1 2.5 Fmo3 Expression (Mean Log Ratio) Supplementary Figure 12. Relation between hepatic Fmo3 expression levels in mice from the C57BL/6J.Apoe- /- and C3H/HeJ Apoe- /- F2 intercross and various phenotypes. The correlations shown are between hepatic expression levels in combined male and female mice (left column), male mice (middle column), and female mice (right column) from the F2 intercross versus aortic sinus atherosclerotic lesion area (top row), HDL cholesterol levels (middle row) and plasma TMAO levels (bottom row). WWW.NATURE.COM/ NATURE | 33 Supplementary Figure 13 Hepatic expression levels of Fmo3 in adult male mice is extremely low. This may have accounted for the lack of correlation with lesion area observed in the microarry study (Supplementary Fig 12, top panel, middle). We therefore performed separate analyses using a more sensitive method to quantify Fmo3 mRNA (qRT-PCR), which is shown below. Note that using the more sensitive and accurate qRT-PCR data, a significant positive correlation between hepatic Fmo3 expression levels and quantitative measures of atherosclerotic plaque burden in males were observed. Male mouse hepatic Fmo3 mRNA expression level relative to Gapdh quantified by RT-PCR and the lowest value was normalized to 1. 100000 r=0.58; p<0.001 Aortic lesion (µm2) 80000 60000 40000 20000 0 0 5 10 15 20 25 30 35 Fmo3 mRNA Supplementary Figure 13. Correlation between hepatic Fmo3 mRNA expression levels and aortic lesion. Male C57BL/6J.Apoe- /- mice at time of weaning (4 weeks) were placed on diets consisting of normal chow or normal chow supplemented with varying amounts of choline or TMAO as indicated in Fig. 3d of the manuscript. Liver was isolated after 16 weeks on the diets at time of aorta recovery for quantification of aortic sinus lesions. Hepatic Fmo3 mRNA expression levels were quantified by RT-PCR. Expression levels are normalized to Gapdh mRNA, and the minimum value was set to 1.0. Aortic lesion area was quantified after staining with Oil-red-O / haematoxylin. WWW.NATURE.COM/ NATURE | 34 Supplementary Figure 14 mouse r=0.10; p=0.09 r=0.19; p=0.002 600000 Aortic lesion (µm2) Aortic lesion (µm2) 600000 300000 300000 0 0 -0.2 0 0.2 Fmo1 Expression (Mean Log Ratio) -0.8 -0.4 0 0.4 Fmo2 Expression (Mean Log Ratio) r=0.35; p<0.001 r=0.12; p=0.04 600000 Aortic lesion (µm2) Aortic lesion (µm2) 600000 300000 0 -2 -1 0 Fmo3 Expression (Mean Log Ratio) 300000 300000 0 0 1 -0.2 r=-0.23; p<0.001 r=-0.01; p=0.87 600000 Aortic lesion (µm2) Aortic lesion (µm2) 600000 300000 0 300000 0 -0.4 -0.2 0 Fmo5 Expression (Mean Log Ratio) 0.2 r=0.001; p=0.98 600000 Aortic lesion (µm2) 0 0.2 Fmo4 Expression (Mean Log Ratio) 300000 0 -0.6 -0.2 0.2 Fmo9 Expression (Mean Log Ratio) 0.6 -0.2 0 Fmo6 Expression (Mean Log Ratio) 0.2 Supplementary Figure 14. Correlation between expression levels of multiple members of the murine family of hepatic flavin monooxygenase (Fmos) and aortic lesion. Hepatic expression levels of the Fmos in t he F2 intercross between atherosclerosis prone C57BL/6J.Apoe- /- and atherosclerosis resistant C3H/HeJ Apoe- /- mice were determined by microarray analysis (described in Methods) and aorta lesions were quantified after staining with Oil-red-O/haematoxylin. WWW.NATURE.COM/ NATURE | 35 Supplementary Figure 15 mouse 200 r=-0.31; p<0.001 100 100 0 0 -0.2 -0.8 0 0.2 Fmo1 Expression (Mean Log Ratio) 200 r=-0.48; p<0.001 HDL Cholesterol (mg/dL) HDL Cholesterol (mg/dL) 200 100 0 200 -1 0 Fmo3 Expression (Mean Log Ratio) r=-0.36; p<0.001 100 1 -0.2 200 r=0.38; p<0.001 HDL Cholesterol (mg/dL) HDL Cholesterol (mg/dL) -0.4 0 0.4 Fmo2 Expression (Mean Log Ratio) 0 -2 100 0 0 0.2 Fmo4 Expression (Mean Log Ratio) r=0.34; p=0.57 100 0 -0.4 200 HDL Cholesterol (mg/dL) r=-0.38; p<0.001 HDL Cholesterol (mg/dL) HDL Cholesterol (mg/dL) 200 -0.2 0 Fmo5 Expression (Mean Log Ratio) 0.2 r=0.012; p=0.84 100 0 -0.6 -0.2 0.2 Fmo9 Expression (Mean Log Ratio) 0.6 -0.2 0 Fmo6 Expression (Mean Log Ratio) 0.2 Supplementary Figure 15. Correlation between expression levels of multiple members of the murine family of hepatic flavin monooxygenase (Fmos) and plasma HDL cholesterol. Hepatic expression levels of the Fmos in the F2 intercross between atherosclerosis prone C57BL/6J.Apoe- /- and atherosclerosis resistant C3H/HeJ Apoe- /- mice were determined by microarray analysis (described in Methods) and plasma HDL cholesterol levels were measured by Cholesterol LiquiColor® Test (Enzymatic) (Stanbio, Boerne, Texas). WWW.NATURE.COM/ NATURE | 36 Supplementary Figure 16 mouse TMAO (µM) 3 0 -0.1 0 0.1 Fmo1 Expression (Mean Log Ratio) 3 TMAO (µM) 0 0 Fmo2 Expression (Mean Log Ratio) 6 0 Fmo4 Expression (Mean Log Ratio) 0.2 r=-0.24; p=0.33 3 0 -0.1 0 Fmo9 Expression (Mean Log Ratio) 0.1 10 3 -2 -0.5 1 Fmo3 Expression (Mean Log Ratio) 2.5 r=-0.27; p=0.29 6 0 -0.2 20 0.5 r=-0.68; p=0.002 6 3 r=0.80; p<0.001 0 -0.5 0.2 6 r=0.27; p=0.28 TMAO (µM) 30 0 -0.2 TMAO (µM) r=0.41; p=0.05 TMAO (µM) TMAO (µM) 6 TMAO (µM) r=-0.028; p=0.91 6 3 0 -0.4 -0.2 0 Fmo5 Expression (Mean Log Ratio) 0.2 -0.1 0 Fmo6 Expression (Mean Log Ratio) Supplementary Figure 16. Correlation between expression levels of multiple members of the murine family of hepatic flavin monooxygenase (Fmos) and plasma levels of TMAO. Hepatic expression levels of the Fmos in t he F2 intercross between atherosclerosis prone C57BL/6J.Apoe- /and atherosclerosis resistant C3H/HeJ Apoe-/- mice were determined by microarray analysis (described in Methods), and plasma TMAO levels were measured using stable isotope dilution LC/MS/MS as described in Methods. WWW.NATURE.COM/ NATURE | 37 0.1 Supplementary Figure 17 Human (female) hepatic Fmos mRNA and plasma levels of TMAO relative to GAPDH 15 r=-0.11 p=0.76 10 TMAO (µM) TMAO (µM) 15 5 0 5 0 0 4 Fmo1 mRNA 8 0 5 Fmo2 mRNA 10 15 TMAO (µM) 15 TMAO (µM) r=-0.31 p=0.28 10 10 r=0.43 p=0.13 10 r=-0.49 p=0.07 5 0 5 0 0 1000 2000 Fmo3 mRNA 3000 0 6 Fmo4 mRNA 12 TMAO (µM) 15 10 r=0.31 p=0.31 5 0 0 200 Fmo5 mRNA 400 Supplementary Figure 17. Correlation between expression levels of multiple members of the human family of hepatic flavin monooxygenases (Fmos) and plasma levels of TMAO. Human liver biopsies and paired samples of fasting blood for plasma isolation were collected. Hepatic expression levels for the Fmos were determined by qRT-PCR in each sample and normalized to the level of Gapdh mRNA in that sample. Plasma levels of TMAO were quantified by stable isotope dilution LC/MS/MS. WWW.NATURE.COM/ NATURE | 38 Supplementary Figure 18 mouse 8 LOD 6 4 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Chromosome Number Supplementary Figure 18. Genome scan results for Fmo3 expression in mice from the C57BL/6J.Apoe- /- and C3H/HeJ.Apoe- /- F2 intercross. LOD score plots for Fmo3 mRNA levels across the 20 autosomal chromosomes in mice using a sex interaction model adjusting for sex, and sex × QTL. The horizontal lines represent the significant (Black, p < 0.05) and suggestive (Red, p < 0.63) LOD thresholds as determined by 1000 permutations. A suggestive ciseQTL for Fmo3 expression (and other Fmo within the Fmo gene cluster) is found on chromosome 1 over the physical location of the Fmo gene cluster. WWW.NATURE.COM/ NATURE | 39 19 20 Supplementary Figure 19 mouse Peritoneal macrophages p=0.075 p=0.22 Cd36 mRNA relative to F4/80 3 p=0.075 2 1 0 chow (n=5) choline TMAO (n=5) (n=4) betaine (n=5) p=0.047 p=0.014 Sr-a1 mRNA relative to F4/80 6 p=0.075 4 2 0 chow (n=5) choline TMAO (n=5) (n=4) betaine (n=5) Supplementary Figure 19. Effect of dietary choline, TMAO and betaine on expression of scavenger receptors by RT-PCR. Female C57BL/6J.Apoe- /mice were placed on diets consisting of normal chow or normal chow supplemented with 1.0% (wt/wt) of choline, 0.12% TMAO or 1.0% betaine at time of weaning (4 weeks). After 1 month, peritoneal macrophages were harvested by lavage, RNA was extracted, and Cd36 and Sr -a1 mRNA levels quantified by RT-PCR. Expression levels are normalized to control F4/80 mRNA for comparison. . WWW.NATURE.COM/ NATURE | 40 Supplementary Figure 20 a b c - + - IgG2b-Alexa Fluor 647 CD11b Fluor 647 IgG2a-Alexa Fluor 647 F4/80 Fluor 647 d F4/80 positive cells f + e CD11b positive cells F4/80 and CD11b positive cells WWW.NATURE.COM/ NATURE | 41 Supplementary Figure 20. Expression of surface scavenger receptor by flow cytometry. Cells sorted with the macrophage specific F4/80 and CD11b double positive staining were used to measure the level of CD36 and SR-A1 protein on the surface of thioglycollate-elicited macrophages in mice. Anti-F4/80 antibody was conjugated with Alexa® 647, anti-CD36 antibody and anti-rat Ig antibody for SR-A1 were conjugated with FITC. a, A typical flow cytometry scattergram to sort cells by forward size channel (FSC) and scattering size channel (SSC). b-c, Histograms of the cells immunostained with rat anti -mouse F4/80 and anti-mouse CD11b and their corresponding isotope controls, rat IgG2a and IgG2b, conjugated with Alexa Fluor® 647, respectively. A dotted black line inside the histograms split cells into F4/80, CD11b positive (+) staining and negative (-) staining ones. d-e, The extracted F4/80 and CD11b positive staining cells, respectively. f, The final extracted double positive cells. WWW.NATURE.COM/ NATURE | 42 Supplementary Figure 21 mouse liver H/E Oil red O MCD control diet choline diet 100 µm Supplementary Figure 21. Representative examples showing haematoxylin/eosin (H/E) and oil-red-O / haematoxylin stainings of liver from mice fed with different levels of choline diet. C57BL/6J.Apoe-/- mice were placed on the different diets at weaning (4 weeks) and sacrificed 16 weeks later. The liver sections were stained with haematoxylin/eosin (H/E) and oil-red-O/haematoxylin. Parallel biochemical analyses on liver tissues for triglyceride and cholesterol content were also performed. Control diet, (0.08±0.01)% (wt/wt) total choline; choline diet, 1.0% (wt/wt) total choline; MCD (methonine and choline deficient diet classically used for induction of steatohepatitis), 0% (wt/wt) total choline. WWW.NATURE.COM/ NATURE | 43 Supplementary Figure 22 mouse aortic root Mouse aorta immunohistochemical staining for macrophages using rat anti mouse F4/80 antibody followed by goat anti rat IgG conjugated with FITC control diet choline diet control diet / Abx choline diet / Abx 200 µm Supplementary Figure 22. Representative immunohistochemical staining of aorta for F4/80 in C57BL/6J.Apoe- /- mouse on chow (control diet, (0.08±0.01)% choline) or choline supplemented diet (choline diet, 1.0% choline) off or on antibiotics (Abx). The aortic root was sectioned and immunohistochemically stained with rat anti mouse F4/80 antibody followed by goat anti rat IgG conjugated with FITC. Section was counterstained with DAPI. WWW.NATURE.COM/ NATURE | 44 Supplementary Figure 23 mouse aortic root Aortic lesion imunohistochemical staining for macrophage scavenger receptor CD36 with anti-CD36 conjugated with FITC and counterstained with DAPI Control Choline Control/Abx Choline/Abx 200 µm Supplementary Figure 23. Representative immunohistochemical staining of aorta for CD36 in C57BL/6J.Apoe- /- mouse on chow (control diet, (0.08±0.01)% choline) or choline supplemented diet (choline diet, 1.0% choline) off or on antibiotics (Abx). The aortic root was sectioned and immunohistochemically stained with rat anti mouse CD36 antibody conjugated with FITC. Section was counterstained with DAPI. WWW.NATURE.COM/ NATURE | 45 Supplementary Figure 24 p=0.94 Total Cholesterol (µg/µg DNA) 4 p=0.01 CH p=0.05 HO CH CH 2 CH 3 2 + CH N CH HO 3 3 CH 3 3 Trimethylamine N-oxide (TMAO) choline 2 3 + CH N CH HO CH 2 0 3 CH 2 C CH 3 CH 3 control choline TMAO DMB (n=10) (n=8) (n=7) (n=11) 3,3-dimethyl-1-butanol (DMB) Supplementary Figure 24. Difference in cholesterol accumulated in macrophages among structurally similar chemical compounds, choline, TMAO and 3,3-dimethyl-1-butanol (DMB). Male C57BL/6J.Apoe- /- mice (15 week old) were placed on normal chow (control) alone or supplemented in the presence of either choline (1.0%), TMAO (0.12%), or DMB (1.0%). Peritoneal macrophages were recovered from the indicated number of mice at 20 weeks of age and cellular cholesterol content was quantified by stable isotope dilution GC/MS, and normalized to DNA content. WWW.NATURE.COM/ NATURE | 46