Metabolomics Profiling of Epicardial Adipose Tissue: MESA and the Rotterdam Study

Abstract

Background

Excess epicardial adipose tissue (EAT) has been associated with cardiovascular diseases such as atrial fibrillation, coronary artery disease, and heart failure. The metabolomic signature of EAT is not well studied.

Methods

Untargeted 1H nuclear magnetic resonance metabolomics profiling of serum was performed (1‐dimensional nuclear magnetic resonance, Carr–Purcell–Meiboom–Gill Echo Train Acquisition, lipidomics) and EAT was measured with computed tomography in MESA (Multi‐Ethnic Study of Atherosclerosis; N=3936) and the Rotterdam study (N=465). Associations between fasting serum metabolites and EAT volume were assessed using cross‐sectional linear regression of individual‐level data in MESA and validated in Rotterdam.

Results

A total of 23 571 metabolomic spectral variables were evaluated. In MESA, after adjustment for age, sex, and race and ethnicity, 38 metabolites were positively and 19 metabolites negatively associated with EAT at a false discovery rate P<0.01. Several metabolites were replicated in Rotterdam, including 1,5‐anhydrosorbitol and N‐acetyl (glycoproteins) that were positively associated with EAT and trimethylamine (phospholipids) that were inversely associated with EAT. Branched‐chain amino acids (leucine, isoleucine, and valine) and 3‐hydroxybutyrate were also associated with EAT in the Rotterdam study. In MESA, apolipoprotein B and very‐low‐density and intermediate‐density lipoprotein fractions were positively associated with EAT and the majority of high‐density lipoprotein subclasses were inversely associated with EAT. Associations were partially attenuated in MESA and fully attenuated in Rotterdam after further adjustment for health and socioeconomic factors.

Conclusions

From >20 000 metabolomic features, 1,5‐anhydrosorbitol, glycoproteins, phospholipids, and atherogenic dyslipidemia markers emerged as significant markers of EAT. Further investigation is warranted to determine whether nuclear magnetic resonance–based metabolic profiling can improve EAT detection with implications for cardiometabolic health.

Nonstandard Abbreviations and Acronyms

EAT

epicardial adipose tissue

MESA

Multi‐Ethnic Study of Atherosclerosis

NEO

Netherlands Epidemiology of Obesity

NMR

nuclear magnetic resonance

Clinical Perspective.

What Is New?

• Using an untargeted metabolomic approach, 1,5‐anhydrosorbitol, glycoproteins, phospholipids, and markers of atherogenic dyslipidemia emerged as strong markers of epicardial adipose tissue.

What Are the Clinical Implications?

• A single, fasting measurement of these metabolites may provide additional information over standard risk markers of ectopic fat (body mass index, fasting glucose, waist circumference, and serum triglycerides).

• Further investigation is warranted to determine whether nuclear magnetic resonance–based metabolic profiling can improve screening and detection of epicardial adipose tissue, as well as monitoring of treatment effects on epicardial adipose tissue, beyond simple anthropometric measures and the hypertriglyceridemic waist to help identify appropriate candidates for interventions and reduce the cardiometabolic complications of visceral and ectopic fat.

Epicardial adipose tissue (EAT) has been associated with occurrence of atrial fibrillation, coronary artery disease, heart failure, and with COVID‐19–related cardiac syndromes. ¹ The absence of a fascia separating left atrial EAT from the underlying left atrial myocardium and a shared blood supply provide a milieu for bidirectional communication between the left atrial EAT and the left atrium, enabling left atrial EAT to serve as an agent for atrial remodeling and development of atrial myopathy, which has been associated with occurrence of atrial fibrillation. EAT is thought to have a role in the development of cardiovascular disease through gene expression, proinflammatory and profibrotic proteins, neuromodulation, and glucose and lipid metabolism. ¹ However, a specific metabolic signature indicating a mechanistic link between EAT and adverse cardiac conditions has not yet been found. ²

Metabolomics profiling has been applied to the study of several cardiometabolic phenotypes, including type 2 diabetes, ³ adiposity, ⁴ and insulin resistance. ⁵ Serum metabolic signatures have been associated with coronary and carotid atherosclerosis. ⁶ Metabolomics profiling has also been used in studies of visceral adipose tissue and other related ectopic fat depots, ⁴ including studies on EAT. Previous metabolomic research on EAT has been done on specific samples of patients, such as those with obesity ⁷ , ⁸ or coronary artery disease ⁹ or in small numbers of apparently healthy individuals. ¹⁰ , ¹¹ To the best of our knowledge, there have been no prior metabolomic studies in large and diverse populations of healthy adults evaluating the metabolic signatures of EAT in unselected individuals without preexisting cardiovascular disease (CVD). In light of a putative link between EAT and CVD through metabolic pathways, we undertook to study a broad range of metabolites associated with EAT volume using a diverse set of contemporary metabolomic profiling methods in 2 large cohort studies of community‐dwelling adults in the United States and the Netherlands.

METHODS

Data‐Sharing Statement

The data that support the findings of this study are available from the respective data coordinating centers of each study after approval of a manuscript proposal.

Study Population and Variable Definitions

Multi‐Ethnic Study of Atherosclerosis

The overall design of the MESA study has been described previously. ¹² Briefly, the MESA study consists of 6814 men and women aged 45 to 84 years who were free of clinical CVD at baseline between 2000 and 2002, of different self‐reported races and ethnicities (White, Black, Chinese American, and Hispanic) enrolled from 6 different sites in the United States. Clinical CVD, leading to exclusion from enrollment, was defined as history of myocardial infarction, angina pectoris, prior revascularization, heart failure, atrial fibrillation, stroke, or peripheral artery disease. Baseline medical history, anthropometric measurements, imaging, and laboratory data for the present study were taken from the first examination of the MESA cohort (July 2000 to August 2002), as previously described. ¹³ Fasting serum samples from 3955 participants randomly selected were collected at the baseline visit to generate untargeted metabolomic profiles in a subset of MESA participants as part of the Development of Combinatorial Biomarkers for Subclinical Atherosclerosis initiative, a collaboration between MESA investigators and scientists at Imperial College London, ⁶ as described below.

Rotterdam Study

The Rotterdam Study is an ongoing prospective population‐based cohort of adults aged ≥40 years living in the Ommoord district in Rotterdam, the Netherlands, with details on study design previously described. ¹⁴ The original cohort (Rotterdam Study‐I) started in 1990, and underwent 3 extensions in 2000 (Rotterdam Study‐II), 2006 (Rotterdam Study‐III), and 2016 (Rotterdam Study‐IV). Participants were examined extensively at study entry, and subsequent follow‐up visits every 3 to 6 years. At the baseline visit, an extensive physical examination was performed, including blood sampling (metabolomic measurements were taken between 1997 and 1999). From 2003 to 2006, all participants who completed a regular visit at the research center were invited to undergo multidetector computed tomography (CT) of the coronary arteries, aortic arch, extracranial carotid arteries, and intracranial carotid arteries, from which EAT was measured. Participants with existing CVD (including atrial fibrillation) at baseline were not excluded in the Rotterdam study. EAT and metabolomic measurements in Rotterdam were performed ≈5 years apart.

For the purposes of the current study, we included 3936 participants from MESA and 465 participants from Rotterdam with available metabolomic and EAT data for analysis. Protocols were approved by the institutional review board at each participating institution for MESA and by the Medical Ethical Committee of the Rotterdam Medical Center for the Rotterdam Study. All participants provided written informed consent.

Metabolomics Measurements

In MESA (the discovery cohort), untargeted ¹H nuclear magnetic resonance (NMR) analysis of serum samples obtained at the baseline examination were performed using a methodology previously described. ¹⁵ The MESA samples used in the current study were analyzed in two phases as part of the EU‐funded Development of Combinatorial Biomarkers for Subclinical Atherosclerosis project. The specific NMR data sets used in the current study include (1) a standard 1‐dimensional NMR spectrum showing resonances from all proton‐containing molecules in the sample, including broad, largely undefined bands from serum proteins, sharper and well‐defined bands from serum lipoproteins (with some classification into their main groups) and 2 sharp peaks from the N‐acetyl group of serum glycoproteins and sharp peaks from a range of small molecule metabolites such as amino acids, simple carbohydrates, organic acids, organic bases and a number of osmolytes; (2) Carr–Purcell–Meiboom–Gill spectrum that attenuates the peaks from the macromolecules and allows better definition of the small molecules; and (3) lipidomics analysis with quantification of lipoprotein subclasses obtained from deconvolution of the methyl peak near δ0.89 of the standard 1‐dimensional NMR spectra using Bruker IVDr Lipoprotein Subclass Analysis B.I.‐LISA (Bruker Biospin, Rheinstetten, Germany), a procedure adapted from the method of Petersen. ¹⁶ Bruker NMR measurements included total high‐density lipoprotein (HDL), low‐density lipoprotein (LDL), triglycerides, and cholesterol as well as analysis of 105 lipoprotein subclasses, including different chemical components of intermediate‐density lipoprotein (density, 1.006–1.019 kg/L), very‐low‐density lipoprotein (VLDL; density, 0.950–1.006 kg/L), LDL (density, 1.09–1.63 kg/L), and HDL (density, 1.063–1.210 kg/L). The LDL subfraction was separated into 6 density classes (LDL‐1, 1.019–1.031 kg/L; LDL‐2, 1.031–1.034 kg/L; LDL‐3, 1.034–1.037 kg/L; LDL‐4, 1.037–1.040 kg/L; LDL‐5, 1.040–1.044 kg/L; LDL‐6, 1.044–1.063 kg/L) and the HDL subfraction into 4 density classes (HDL‐1, 1.063–1.100 kg/L; HDL‐2, 1.100–1.125 kg/L; HDL‐3, 1.125–1.175 kg/L; HDL‐4, 1.175–1.210 kg/L). These specific NMR spectra have been previously tested for quality control, harmonization, and alignment. ¹⁷ Metabolomics analysis was identical in the Rotterdam Study (the replication cohort), except that lipoprotein subclass analysis was not performed and was available only in the MESA cohort.

Epicardial Adipose Tissue Measurements

The terms epicardial fat and pericardial fat have often been used interchangeably throughout the literature, although they can be distinguished anatomically. ¹⁸ Fat related to the pericardium consists of 2 layers: the visceral, epicardial fat layer (which is enclosed between the myocardium and the visceral pericardium) and the parietal, pericardial fat layer (external to the parietal pericardium and surrounding the cardiac silhouette). In both the MESA and Rotterdam studies, it was difficult to separately quantify the epicardial and pericardial fat due to lack of consistently clear pericardial definition using the available imaging, especially in lean individuals. To address this issue, epicardial fat was measured in a random sample of 159 MESA participants, and the Spearman correlation coefficient between pericardial and epicardial fat was 0.92 (P<0.0001). ¹⁹ Therefore, in light of the high correlation between epicardial and pericardial fat, the total pericardial fat volume measurement was used and is collectively referred to as EAT for this analysis. At the first MESA examination, CT scans of the heart were performed either with an ECG‐triggered electron‐beam scanner or with prospectively ECG‐triggered scan acquisition with a multidetector CT system that acquired 4 simultaneous 2.5‐mm slices for each cardiac cycle in a sequential or axial scan mode. Three experienced CT analysts measured EAT volume on the obtained images of the heart at a single CT reading center. ¹⁹ Slices within 15 mm above and 30 mm below the superior extent of the left main coronary artery were included; this region of the heart was selected because it includes the pericardial fat located around the proximal coronary arteries (left main coronary, left anterior descending, right coronary, and circumflex arteries). The anterior border of the volume was defined by the chest wall and the posterior border by the aorta and the bronchus. Volume analysis software (GE Health Care, Waukesha, WI) was used to discern fat from other tissues with a threshold of −190 to −30 Hounsfield units. The volume was the sum of all voxels containing fat. This method of measurement of EAT volume was highly correlated with total pericardial fat volume ²⁰ in a random subset of 10 Diabetes Heart Study participants (correlation coefficient, 0.93; P<0.0001). To assess reproducibility, a random sample of 80 MESA participants was selected, and their CT scans were reread. The intraclass correlation coefficients of intra‐ and interreader reliability were 0.99 and 0.89, respectively, for EAT.

In Rotterdam, a 16‐slice or 64‐slice multidetector CT scanner was used to perform nonenhanced CT scanning. Detailed information regarding imaging parameters of all scans has been reported. ²¹ Quantification of EAT volume was performed from the cardiac CT scan using a fully automatic quantification tool. This quantification method consisted of 2 steps: (1) whole heart segmentation and (2) EAT volume quantification. Analysts used the obtained whole heart segmentation as a region of interest and a threshold window of −30 to −200 Hounsfield units for the quantification of the amount of fat. ²² A connected‐component analysis was applied to all adipose tissue voxels using an 18‐neighborhood rule, to remove regions <10 voxels (2.8 mm³) in size, which was considered to be noise. This fully automatic method was validated by an expert reviewer panel and proved to be as good as manual segmentation. ²³

Covariates

Through clinical examination, questionnaires, interviews, and linkage to health records, extensive information on sociodemographic, lifestyle, disease history, examination, and laboratory data were collected. Covariates included age, self‐reported sex, race, body mass index (BMI), waist circumference, smoking status (current, former, never), alcohol use status (current, former, never), fasting blood glucose, HDL cholesterol, total cholesterol, triglycerides, systolic blood pressure, physical activity (as defined per cohort), diabetes status, lipid‐lowering medication, blood pressure–lowering medication, and history of CVD (including coronary heart disease, stroke, and heart failure). Education level was determined from a self‐report questionnaire. Physical activity was derived using a self‐reported frequency and type of leisure time physical activity and a standard conversion for metabolic equivalence units. ²⁴ Weight and height were measured using a balance‐beam scale and stadiometer, respectively, and used to calculate BMI as weight (kilograms) divided by height (meters) squared. Waist circumference was measured using a steel measuring tape (standard 4‐oz tension) from midway between the last rib and the iliac crest at normal breathing.

Statistical Analysis

All analyses were performed by the statistician for each cohort, separately, by following the same analytic protocol. For all the analyses, to allow for comparisons, metabolites were natural log‐transformed and standardized to a mean±SD of 0±1. EAT was confirmed to be approximately normally distributed. Baseline characteristics of participants were summarized as frequencies and percentages for categorical variables and as means±SDs for continuous variables. In the Rotterdam Study, the maximum missing rate for covariates was up to 3.4% of the participants. Missing values in covariates were imputed using the multiple imputation method with the “mice” package in R (R Foundation for Statistical Computing, Vienna, Austria). Ten imputed data sets were generated. For each metabolite in EAT analyses, multivariable linear regression models were constructed to assess the associations of metabolites with EAT for NMR, Carr–Purcell–Meiboom–Gill Echo Train Acquisition, and lipidomics (MESA only) experiments separately. Metabolites were set as the exposure variables, and EAT was set as the outcome variable on the basis of a hypothesis‐free design since the biology of metabolomics and EAT may be bidirectional (ie, metabolites may influence EAT accumulation/function and EAT accumulation may influence downstream metabolic processes). Analyses were adjusted for potential confounding using the following 2 models: Model 1 was adjusted for age, sex, and race and ethnicity; model 2 was additionally adjusted for education level, BMI, waist circumference, smoking status (current, former, never), alcohol use status (current, former, never), fasting blood glucose, HDL cholesterol, total cholesterol, triglycerides, systolic blood pressure, physical activity (as defined per cohort), diabetes status, lipid‐lowering medication, and blood pressure–lowering medication. Analyses in the Rotterdam Study were additionally adjusted for history of CVD. Models for lipids/lipoproteins in MESA were adjusted only for age, sex, and race and ethnicity to avoid overfitting of the model (given strong correlations between lipids/lipoproteins and other risk factors and medication use). To account for multiple testing, P values were adjusted for false discovery rate using the Benjamin–Hochberg method; P values <0.01 were considered statistically significant. To visualize the metabolomic associations, adjusted P values were log₁₀‐transformed and multiplied by the direction of change (the sign of the β estimate or coefficient) to generate Manhattan plots with significant features color‐coded on the Manhattan and median spectrum plots as red (positive variation) and blue (negative variation). Statistical analyses were performed using SAS version 9.4 software (SAS Corporation, Cary, NC) and Stata Statistical Software version 14.0 (Statacorp, College Station, TX) and R software version 4.2.2.

RESULTS

Characteristics of the discovery (MESA) and replication (Rotterdam) study cohorts are presented in Table 1. Both cohorts were primarily middle‐aged and older adults, with ≈50% women. The MESA cohort was racially and ethnically diverse, with ≈60% non‐White participants, compared with the Rotterdam Study, which was predominantly White. The median BMIs, waist circumferences, and EAT volumes were modestly higher in MESA than in the Rotterdam Study, generally reflecting known demographic and anthropometric differences between the United States and the Netherlands.

Table 1.

Baseline^* Characteristics of the MESA and Rotterdam Cohorts

	MESA N=3936	Rotterdam study N=465	P value
Age, y, mean±SD	62.9±10.3	68.9±5.2	<0.0001
Female sex, n (%)	1995 (50.7)	218 (46.9)	0.12
Race and ethnicity, n (%)
White	1524 (38.7)	458 (98.5)	<0.0001
Black	965 (24.5)
Hispanic	922 (23.4)
Chinese	525 (13.3)
Education, n (%)
MESA‐specific
No schooling	43 (1.1)
Grades 1–8	416 (10.6)
Grades 9–11	284 (7.2)
Completed high school/GED	719 (18.3)
Some college but no degree	634 (16.1)
Technical school certificate	278 (7.1)
Associate degree	187 (4.8)
Bachelor's degree	678 (17.2)
Graduate or professional school	680 (17.3)
Rotterdam‐specific
Primary		49 (10.5)
Lower/intermediate general education		173 (37.2)
Intermediate vocational education		153 (32.9)
Vocational education or university education		90 (19.4)
Type 2 diabetes, n (%)	398 (10.1)	54 (11.6)	0.31
History of CVD	0 (0)	66 (14.2)	<0.0001
Body mass index, kg/m2, mean±SD	28.2±5.4	26.9±4.0	<0.0001
Waist circumference, cm, mean±SD	98.0±14.1	93.7±10.6	<0.0001
Smoking, n (%)			<0.0001
Never	1980 (50.3)	140 (30.1)
Former	1457 (37.0)	253 (54.4)
Current	483 (12.3)	72 (15.5)
Alcohol drinking, n (%)			<0.0001
Never	848 (21.5)	21 (4.5)
Former	923 (23.5)	16 (3.4)
Current	2132 (54.2)	428 (92.0)
Systolic blood pressure, mm Hg, mean±SD	127.0±21.3	144.0±20.8	<0.0001
Blood pressure–lowering medication, n (%)	1327 (33.7)	162 (34.8)	0.63
Total cholesterol, mg/dL, mean±SD	194.6±36.1	223.1±37.0	<0.0001
HDL cholesterol, mg/dL, mean±SD	50.7±14.7	54.0±14.9	<0.0001
Triglycerides, mg/dL, mean±SD	135.0±93.2	134.7±65.5	0.93
Lipid lowering medication, n (%)	657 (16.7)	83 (17.8)	0.53
Fasting blood glucose, mg/dL, mean±SD	98.3±31.1	105.2±22.1	<0.0001
Antihyperglycemic medication, n (%)	358 (9.1)
Physical activity, n (%)
Light	1059 (26.9)
Moderate	1847 (46.9)
Vigorous	1030 (26.2)
Physical activity, MET‐h/week, mean±SD or median (IQR)	64.5±89.0	83.8 (61.0–112.9)	†
Epicardial adipose tissue (mL)	80.7±41.8	113.5±45.5	<0.0001

Data represent mean±SD or number (proportion), respectively. CVD indicates cardiovascular disease; GED, General Educational Development; HDL, high‐density lipoprotein; IQR, interquartile range; MESA, Multi‐Ethnic Study of Atherosclerosis; MET, metabolic equivalents.

^*Baseline refers to the date that the blood sample used for metabolomics profiling was obtained. Gray‐shaded cells denote when a specific variable is not available in the particular cohort study.

^† P value cannot be computed due to different measures of variability used.

In both MESA and Rotterdam, there were significant associations between metabolites and EAT; 38 metabolites were positively and 19 metabolites negatively associated with EAT at a false discovery rate P<0.01, including proteins, small molecules, and lipids. Tables S1 and S2 each show a summary of the metabolites associated with EAT (both positive and negative associations) in MESA and Rotterdam, respectively. Table 2 shows the summary of the most significant associations between metabolites and EAT in the fully adjusted models for the 1‐dimensional NMR and Carr–Purcell–Meiboom–Gill Echo Train Acquisition for MESA and age, sex, and race and ethnicity adjusted models for the 1‐dimensional NMR and Carr–Purcell–Meiboom–Gill Echo Train Acquisition for Rotterdam, and age, sex, and race and ethnicity adjusted models for lipid subclasses experiments (only in MESA as these were not available in Rotterdam) (see Tables S3 and S4 for individual effect estimates). There were no significant associations of metabolites with EAT in the fully adjusted model in Rotterdam. Representative median NMR spectra and Manhattan and Forest plots demonstrating those associations are shown in the Figure. The metabolites found to be associated with EAT in adjusted models in both cohorts included 1,5‐anhydrosorbitol and N‐acetyl (glycoproteins), which were positively associated with EAT, and trimethylamine (phospholipid), which was inversely associated with EAT. Other metabolites, including branched‐chain amino acids (leucine, isoleucine, and valine) and 3‐hydroxybutyrate, showed associations (both positive and negative) in the Rotterdam Study (Table 2). In the lipidomics analysis, apolipoprotein B and VLDL and intermediate‐density lipoprotein fractions were positively associated with EAT, whereas the majority of HDL subclasses were inversely associated with EAT (Table 2). When lipid/lipoprotein associations were additionally adjusted for BMI and other traditional cardiovascular risk factors, the associations with EAT were significantly attenuated such that only HDL‐4 apolipoprotein A1 and HDL‐4 cholesterol (positively associated) and HDL‐1 apolipoprotein A2, apolipoprotein A1, phospholipids, and triglycerides (negatively associated) remained significantly associated at the false discovery rate level of <0.01.

Table 2.

Summary of Significant Metabolites Associated With EAT in MESA and Rotterdam in 1‐Dimensional NMR and CPMG Experiments

Experimental model	MESA		Rotterdam Study
	Positive associations	Negative associations	Positive associations	Negative associations
1‐dimensional NMR	1,5‐Anhydrosorbitol l‐Threonine Glycerol CH2CH2C=C (fatty acyl chains) C=CHCH2HC=C (fatty acyl chains) NCH3(N‐acetyl groups)	Trimethylamine (choline head group, phospholipids)	l‐Lactate (CH2)n; CH3CH2 (CH2)n; CH3CH2CH2 and CH2 (fatty acyl chains) NCH3 (N‐acetyl groups) CH2CO (fatty acyl chains) 1,5‐Anhydrosorbitol Glycerol l‐Threonine d‐Glucose CH2OCOR (glyceryl moiety) CHOCOR (glyceryl moiety) l‐Tyrosine	CH3 (C18 from cholesterol) CH3 (C26 and C27 from cholesterol); CH3(CH2)n (fatty acyl chains) 3‐hydroxyisobutyrate CH2CH2C=C (fatty acyl chains) l‐Lysine Acetate l‐Glutamine Pyroglutamate Citrate CH2 (lysyl groups) Trimethylamine (choline head group) 3‐Hydroxybutyrate l‐Histidine NH (protein side chains)
CPMG	l‐Glutamate NCH3 (N‐acetyl groups)	CH3 (C26 and C27 from cholesterol) CH3(CH2)n (fatty acyl chains) CH3CH2CH2C=(fatty acyl chains) (CH2)n, CH3CH2 (CH2) n and CH3CH2CH2 (fatty acyl chains) CH2CH2CO (fatty acyl chains) CH2C=C (fatty acyl chains) CH2CO (fatty acyl chains) Trimethylamine (choline head group, Phospholipids) CH=CH (fatty acyl residues)	l‐Lactate l‐Leucine l‐Valine Creatinine l‐Arginine l‐Proline l‐Isoleucine d‐mannose d‐Glucose CH3 (C18 from Cholesterol) CH=CH (fatty acyl residues) CH3CH2CH2C (fatty acyl chains); CH3 (C26 and C27 from cholesterol) (CH2)n (fatty acyl chains) CH3CH2(CH2)n; (CH2)n and CH3CH2CH2 (fatty acyl chains) CHOCOR (glyceryl moiety) CH2OCOR (glyceryl moiety) CH2CH2CO (fatty acyl chains) CH2CO (fatty acyl chains) CH3 (C21 from cholesterol)	CH2 (lysyl groups) l‐Asparagine 1,5‐anhydrosorbitol Choline N(CH3)3 (choline head group) Proline betaine Pyroglutamate Myo‐inositol l‐Lysine N,N‐dimethylglycine Pyroglutamate 3‐Hydroxybutyrate 3‐Methylhistidine Formate l‐Ornithine l‐Glutamine Creatine

Model adjusted for age, sex, race and ethnicity, education level, body mass index, waist circumference, smoking status (current, former, never), alcohol use status (current, former, never), fasting blood glucose, high‐density lipoprotein cholesterol, total cholesterol, triglyceride level, systolic blood pressure, physical activity, diabetes status, lipid‐lowering medication, and blood pressure–lowering medication in MESA and age, sex, race and ethnicity in Rotterdam. Effect estimate represents the difference in EAT volume (mL) per 1 SD in metabolite intensity (relative units). Significance level is false discovery rate–corrected P value <0.01. CPMG indicates Carr–Purcell–Meiboom–Gill; EAT, epicardial adipose tissue; MESA, Multi‐Ethnic Study of Atherosclerosis; and NMR, nuclear magnetic resonance.

Figure 1. Representative median NMR spectra and Manhattan and Forrest plots demonstrating associations between metabolites and EAT for the 1‐dimensional NMR, CPMG in MESA (top panel) and Rotterdam Study (bottom panel), and lipid subclasses experiments in MESA (side panel).

Model adjusted for age, sex, race and ethnicity, education level, body mass index, waist circumference, smoking status (current, former, never), alcohol use status (current, former, never), fasting blood glucose, high‐density lipoprotein cholesterol, total cholesterol, triglyceride level, systolic blood pressure, physical activity, diabetes status, lipid‐lowering medication, and blood pressure–lowering medication in MESA and age, sex, and race and ethnicity in Rotterdam. Model for lipids/lipoproteins adjusted for age, sex, and race and ethnicity. Adjusted P values were log₁₀‐transformed and multiplied by the direction of change (the sign of the β estimate or coefficient) to generate Manhattan plots with significant features color‐coded on the Manhattan and median spectrum plots as red (positive variation) and blue (negative variation). Underlined: NMR‐visible protons corresponding to the identified molecules. CPMG indicates Carr–Purcell–Meiboom–Gill; EAT, epicardial adipose tissue; MESA, Multi‐Ethnic Study of Atherosclerosis; and NMR, nuclear magnetic resonance.

DISCUSSION

Using an untargeted metabolomic platform in a large, multiethnic population cohort (MESA), we identified a metabolite signature associated with EAT linked to several putative biological pathways, including a marker of glycemic control, glycoproteins, phospholipids, and lipids/lipoproteins. We then sought to replicate our findings in a separate epidemiological cohort (Rotterdam Study) and found that 1,5‐anhydrosorbitol and glycoproteins (N‐acetyl groups, positively associated) and phospholipids (trimethylamine head groups) remained associated with EAT and 3‐hydroxybutyrate was found to be inversely associated in Rotterdam, suggesting that a single, fasting measurement of metabolites can provide biological information beyond standard risk markers of epicardial adipose tissue. We believe these findings provide insight into potential mechanisms underpinning EAT metabolism distinct from generalized obesity and help to define a metabolic signature of epicardial adiposity.

1,5‐Anhydrosorbitol was observed to be positively associated with EAT in both MESA and Rotterdam studies. This metabolite is a validated marker of short‐term glycemic metabolism ²⁵ and has been associated with coronary artery calcification. ²⁶ It is excreted in the urine when its level exceeds the renal threshold. It is reabsorbed in the renal tubules and is competitively inhibited by glucosuria, which leads to a reduction in its level in serum. Due to these features, 1,5‐anhydrosorbitol was recently proposed as biomarker of the effectiveness and adherence to the treatment with sodium–glucose cotransporter‐2 inhibitors. ²⁷ Given the established link between visceral/ectopic body fat and prediabetes and type 2 diabetes, ²⁸ the observation that 1,5‐anhydrosorbitol is associated with EAT suggests that the pathophysiology of EAT extends beyond the local cardiac environment and may influence even short‐term glycemic excursions and systemic glucose control. Interestingly, EAT displays a unique transcriptome in patients with diabetes, as demonstrated by Camarena and colleagues. ²⁹

We also found that glycoproteins were significantly positively associated with EAT in both the MESA and Rotterdam studies. Glycoproteins may perform a variety of cellular functions, including enzymatic catalysis, protein folding, conformation, and stabilization of biological membranes important for metabolic homeostasis ³⁰ ; perturbation of this highly regulated system may increase circulating concentrations of glycoproteins and represent a potential biomarker of visceral adiposity‐related disease risk. ³¹ Indeed, Neeland has previously demonstrated that glycoproteins were positively associated with visceral adipose tissue in the MESA and NEO (Netherlands Epidemiology of Obesity) studies. ⁴ Based on prior observations that visceral adipose tissue and EAT are highly correlated with all markers of cardiometabolic risk, ¹¹ it is not surprising that glycoproteins were also associated with EAT in the current study. Notably, both glycoproteins and EAT are associated with inflammation and early stages of atherosclerosis, independently of obesity. ³² , ³³

In the Rotterdam Study, 3‐hydroxybutyrate (or β‐hydroxybutyrate) was found to be negatively associated with EAT, despite the relatively small sample size. While replication is needed, the findings seem relevant. 3‐Hydroxybutyrate is a ketone body with several purportedly beneficial effects in the heart, whose production can be enhanced by sodium–glucose cotransporter‐2 inhibitors and other cardiometabolic medications. ³⁴ Importantly, 3‐hydroxybutyrate has been shown to act as an antagonist of free fatty acid receptor‐3, a G protein–coupled receptor activated by butyrate and other short‐chain free fatty acids. ³⁵ Free fatty acid receptor‐3 promotes adipogenesis via induction of adipokines (eg, adiponectin, leptin) synthesis and secretion in preadipocytes and in differentiated adipocytes. ³⁵ It is thus plausible that 3‐hydroxybutyrate may act as an endogenous suppressor of EAT growth/volume, which could explain its negative association with EAT observed in our study. Notably, Malavazos and colleagues recently showed that activation of EAT glucose‐dependent insulinotropic receptor is negatively associated with genes involved in degradation of ketone bodies, such as BDH2 and HMCGSI, ³⁶ whereas EAT glucagon‐like peptide 1 receptor is positively associated with genes enhancing ketones synthesis such as OXCT2. ³⁷ Studies have also shown that EAT glucagon‐like peptide 1 receptor and glucose‐dependent insulinotropic receptor are associated with increased free fatty acids oxidation by the adjacent myocardium. Targeting EAT with dual glucagon‐like peptide 1–gastric inhibitory polypeptide analogues and with sodium–glucose cotransporter‐2 inhibitors can improve free fatty acids use and modulate ketone bodies availability as alternate fuel to the adjacent myocardium. ³⁸ EAT can therefore serve as mediator for the use of lipid‐derived energy.

For the first time, we found a positive correlation between EAT and branched‐chain amino acidss, such as leucine, isoleucine, and valine, although only in the Rotterdam cohort. Recent evidence suggests multiple effects of dietary and endogenously synthesized amino acids in adipose tissue thermogenesis and metabolism. ³⁹ Under physiological conditions, EAT serves as source of energy and thermoregulation for the heart. EAT brown‐fat function declines with aging and pathological conditions. Interestingly, cardiometabolic drugs such as the glucagon‐like peptide 1–gastric inhibitory polypeptide analogues can target and promote thermogenic‐like amino acid profiles in fat depots like EAT. While results are still preliminary, enhanced EAT thermogenic activity may produce beneficial cardiac effects.

It is well known that both visceral adipose tissue and EAT are associated with an atherogenic, dyslipidemic lipid/lipoprotein profile, including high triglycerides, low HDL‐C, ⁴⁰ , ⁴¹ smaller LDL and HDL particle size, larger VLDL size, and increased LDL and VLDL particle number. ⁴ , ⁴² , ⁴³ Indeed, in our study, HDL‐related particles were inversely associated with EAT, whereas LDL‐, triglyceride‐ and VLDL‐related particles were consistently positively associated with EAT in MESA. Abnormalities in triglycerides and VLDL are more closely linked with entities classically related to visceral and ectopic fat, such as the metabolic syndrome, insulin resistance, and the hypertriglyceridemic waist, ⁴⁴ , ⁴⁵ whereas alterations in HDL metabolism likely relate to atherogenesis through different mechanisms. ⁴⁶ , ⁴⁷ Therefore, our results may reflect multiple mechanistic pathways through which EAT and lipids/lipoproteins interact to influence cardiovascular and metabolic risk. It should be noted that adjustment for BMI attenuated many of the associations in the lipidomics analysis, suggesting that the associations between metabolites and EAT may at least in part be confounded by generalized adiposity (eg, BMI). The lipidomic profile of EAT is of particular interest for the clinical implications, also in asymptomatic and apparently health subjects. ⁷ Given its distinctive transcriptome ⁴⁸ and unobstructed proximity to the myocardium and coronary arteries, EAT can serve as a local source of ectopic lipids. ⁴⁹ Epicardial adipocyte lipotoxicity can contribute to the lipid buildup in the coronary arteries and myocardial disarray, leading to coronary artery disease, heart failure, and atrial fibrillation.

A growing number of studies have used metabolic profiling as a tool for biomarker discovery in body fat distribution, including EAT, albeit most studies were relatively small in sample size and performed in populations with existing CVDs. De Larochelliere and colleagues demonstrated that even in apparently healthy young men and women without obesity that accumulation of EAT was associated with a worse cardiometabolic profile regardless of BMI. ¹¹ Similarly, Scherer and colleagues showed that diacylglycerols and phosphoglycerols were specifically associated with EAT in a cohort of 40 healthy women with obesity. ⁷ Other studies have reported on associations between EAT and metabolites in prevalent cardiac disorders such as coronary artery disease ⁹ and heart failure. ⁵⁰ The majority of prior studies were performed in a single cohort without any external validation or replication. One of the strengths of our investigation is the use of 2 well‐characterized prospective cohorts, 1 for derivation and 1 for replication, each with dedicated imaging assessments of EAT, rather than relying on surrogate markers, such as anthropometric measurements. Furthermore, we use robust untargeted NMR‐based experiments initially to broadly characterize the metabolic phenotype related to EAT and then replicate our findings using a similar approach in a cohort that is well diversified demographically and geographically from the derivation cohort. All individuals in our study had assessments of BMI, waist circumference, and fasting glucose and triglycerides, allowing us to adjust for overall adiposity, glucose intolerance, and dyslipidemia.

Several limitations of the study merit comment. First, our findings should be primarily understood within a biological context; the utility of these metabolites for use in predictive modeling when added to standard clinical risk scores requires further study. Second, slightly different imaging methods were used to estimate EAT in each cohort; however, each method has been validated and is reproducible. Third, although the cohorts varied both geographically and demographically, the replication observed across cohorts despite these differences in study populations (different amounts of EAT and different demographics) makes our findings robust. The sample size was much smaller in Rotterdam, reducing power to detect significant associations, which may explain, at least in part, the replication of several metabolites when adjusted for age, sex, and race and ethnicity, but the attenuation of these associations and lack of replication observed across cohorts when fully adjusted for all covariates in Rotterdam. It is also possible that differences in ethnicity, diet, or distribution of obesity between the cohorts could partially explain the variability observed in metabolite associations. Furthermore, these differences may at least partially explain the observation that some metabolites found to be significant in MESA are not replicated in the Rotterdam study. In that context, our findings may not be generalizable to more diverse populations. Additionally, the analytical strategy using separate linear regression models for each metabolite might have missed potential nonlinear relationships between metabolites and EAT, as well as potential interaction effects among metabolites. It should also be noted that the lipid/lipoprotein observations were limited to the MESA cohort and were not assessed for replication in Rotterdam, as these assays were not performed in the Rotterdam Study. Fourth, we cannot generalize to other populations not well represented in either cohort in which alternative metabolite relationships may exist. Fifth, because our study was cross‐sectional by design, we cannot comment on the relationship between temporal changes in metabolite levels and epicardial fat. Sixth, we cannot exclude the possibility of residual confounding (eg, kidney function, diet, and medication impact) or time‐varying confounding given the time gap between metabolomic measurements and EAT measurements in Rotterdam, which might explain some of the key metabolomic differences between the 2 cohorts.

In conclusion, using an untargeted metabolomics approach, 1,5‐anhydrosorbitol, glycoproteins, phospholipids, and markers of atherogenic dyslipidemia emerged as strong markers of EAT. A single, fasting measurement of these metabolites may provide additional information over standard risk markers of ectopic fat (BMI, fasting glucose, waist circumference, and serum triglycerides). Further investigation is warranted to determine whether NMR‐based metabolic profiling can improve screening and detection of EAT, as well as monitoring of treatment effects on EAT, beyond simple anthropometric measures and the hypertriglyceridemic waist to help identify appropriate candidates for interventions and reduce the cardiometabolic complications of visceral and ectopic fat.

Sources of Funding

This research was supported by grant R01HL155718 from the National Heart, Lung, and Blood Institute, contracts 75N92020D00001, HHSN268201500003I, N01‐HC‐95159, 75N92020D00005, N01‐HC‐95160, 75N92020D00002, N01‐HC‐95161, 75N92020D00003, N01‐HC‐95162, 75N92020D00006, N01‐HC‐95163, 75N92020D00004, N01‐HC‐95164, 75N92020D00007, N01‐HC‐95165, N01‐HC‐95166, N01‐HC‐95167, N01‐HC‐95168 and N01‐HC‐95169 from the National Heart, Lung, and Blood Institute, and by grants R01HL085323 from the National Heart, Lung, and Blood Institute, and UL1‐TR‐000040, UL1‐TR‐001079, and UL1‐TR‐001420 from the National Center for Advancing Translational Sciences. The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa‐nhlbi.org. This paper has been reviewed and approved by the MESA Publications and Presentations Committee. The Development of Combinatorial Biomarkers for Subclinical Atherosclerosis project was supported by a grant from the European Union Seventh Framework Programme (305422). The Rotterdam Study is supported by Erasmus Medical Center and Erasmus University Rotterdam; the Netherlands Organization for Health Research and Development (ZonMw); the Research Institute for Diseases in the Elderly (RIDE); the Ministry of Education, Culture and Science; the Ministry of Health, Welfare and Sports; the European Commission (DG XII); and the Municipality of Rotterdam. This manuscript is part of the Stratification of Obesity Phenotypes to Optimize Future Obesity Therapy project, which has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement number 875534. This joint undertaking support from the European Union's Horizon 2020 research and innovation program and European Federation of Pharmaceutical Industries and Associations and Type 1 Diabetes Exchange, Juvenile Diabetes Research Foundation, and Obesity Action Coalition.

Disclosures

Dr Neeland has received honoraria, consulting, and speaker's bureau fees from Boehringer Ingelheim/Lilly Alliance; consulting fees from Novo Nordisk; travel support from Nestle Health Science; and consulting fees from AMRA Medical. Dr Graca is now an employee of the Syngenta group. This manuscript is part of the Stratification of Obesity Phenotypes to Optimize Future Therapy project, which has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement number 875534. This joint undertaking received support from the European Union's Horizon 2020 research and innovation program and European Federation of Pharmaceutical Industries and Associations and Type 1 Diabetes Exchange, Juvenile Diabetes Research Foundation, and Obesity Action Coalition. The communication reflects the author's view and neither the Innovative Medicines Initiative nor the European Union, European Federation of Pharmaceutical Industries and Associations, or any associated partners are responsible for any use that may be made of the information contained therein. The remaining authors have no disclosures to report.

Supporting information

Tables S1–S4