Cardiovascular Biomarkers of Obesity and Overlap With Cardiometabolic Dysfunction

Abstract

Background

Obesity may be associated with a range of cardiometabolic manifestations. We hypothesized that proteomic profiling may provide insights into the biological pathways that contribute to various obesity‐associated cardiometabolic traits. We sought to identify proteomic signatures of obesity and examine overlap with related cardiometabolic traits, including abdominal adiposity, insulin resistance, and adipose depots.

Methods and Results

We measured 71 circulating cardiovascular disease protein biomarkers in 6981 participants (54% women; mean age, 49 years). We examined the associations of obesity, computed tomography measures of adiposity, cardiometabolic traits, and incident metabolic syndrome with biomarkers using multivariable regression models. Of the 71 biomarkers examined, 45 were significantly associated with obesity, of which 32 were positively associated and 13 were negatively associated with obesity (false discovery rate q<0.05 for all). There was significant overlap of biomarker profiles of obesity and cardiometabolic traits, but 23 biomarkers, including melanoma cell adhesion molecule (MCAM), growth differentiation factor‐15 (GDF15), and lipoprotein(a) (LPA) were unique to metabolic traits only. Using hierarchical clustering, we found that the protein biomarkers clustered along 3 main trait axes: adipose, metabolic, and lipid traits. In longitudinal analyses, 6 biomarkers were significantly associated with incident metabolic syndrome: apolipoprotein B (apoB), insulin‐like growth factor‐binding protein 2 (IGFBP2), plasma kallikrein (KLKB1), complement C2 (C2), fibrinogen (FBN), and N‐terminal pro‐B‐type natriuretic peptide (NT‐proBNP); false discovery rate q<0.05 for all.

Conclusions

We found that the proteomic architecture of obesity overlaps considerably with associated cardiometabolic traits, implying shared pathways. Despite overlap, hierarchical clustering of proteomic profiles identified 3 distinct clusters of cardiometabolic traits: adipose, metabolic, and lipid. Further exploration of these novel protein targets and associated pathways may provide insight into the mechanisms responsible for the progression from obesity to cardiometabolic disease.

Nonstandard Abbreviations and Acronyms

ADM

Adrenomedullin

AGP1

Alpha‐1 acid glycoprotein

ANGPTL3

Angiopoietin‐like 3

apoB

apolipoprotein B

B2M

Beta‐2‐microglobulin

BCHE

Butyrylcholine esterase

BIKUNIN

AMBP‐bikunin

C2

Complement C2

CD14

Monocyte differentiation antigen

CD163

Cluster of differentiation 163

CD40L

Soluble CD40 ligand

CD56 or NCAM

Neural cell adhesion molecule

CD5L

CD5 antigen‐like

CNTN1

Contactin 1

CRP

C‐reactive protein

CXCL16

Chemokine (C‐X‐C motif) ligand 16

DM

diabetes mellitus

EFEMP1

EGF containing fibulin‐like extracellular matrix protein 1

FBN

Fibrinogen

FG

fasting glucose

FHS

Framingham Heart Study

GDF15

growth differentiation factor‐15

GMP140

Granule membrane protein 140

HOMA‐IR

Homeostatic Model Assessment of Insulin Resistance

IGF‐1

Insulin‐like growth factor 1

IGFBP1

Insulin‐like growth factor binding protein 1

IGFBP2

Insulin‐like growth factor binding protein 2

KLKB1

Plasma kallikrein

LPA

lipoprotein(a)

MCAM

melanoma cell adhesion molecule

MCP1

Monocyte chemotactic molecule 1

MetS

metabolic syndrome

MMP8

Matrix metallopeptidase 8

MMP9

Matrix metallopeptidase 9

MPO

Myeloperoxidase

NT‐proBNP

N‐terminal pro‐B‐type natriuretic peptide

OSTEO

Osteocalcin

PAI1

Plasminogen activator inhibitor 1

PC1

principal component 1

PC2

principal component 2

REG1A

Lithostathine‐1‐alpha

SAA1

Serum amyloid A1

SAT

subcutaneous adipose tissue

SDF1

Stromal cell‐derived factor 1

sICAM1

Intercellular adhesion molecule 1

SRAGE

Receptor for advanced glycation endproducts

TIMP1

Tissue inhibitor of metalloproteinases 1

UCMGP

Uncarboxylated matrix Gla protein

VAT

visceral adipose tissue

VEGF

Vascular endothelial growth factor

WC

waist circumference

What Is New?

• In an analysis of 71 circulating cardiovascular disease–related protein biomarkers ascertained in 6981 participants free of cardiovascular disease, we demonstrate abundant associations of biomarkers with obesity, cardiometabolic traits, and computed tomography measures of regional adiposity.

• Although all biomarkers significantly associated with obesity were also associated with 1 or more cardiometabolic traits, several biomarkers were unique to metabolic traits only.

• Clustering based on biomarker profiles revealed 3 distinct groups of traits: (1) adipose, (2) metabolic, and (3) lipid. The biomarkers apoB (apolipoprotein B), KLKB1 (plasma kallikrein), C2 (complement C2), and FBN (fibrinogen) were associated with an increased risk of future metabolic syndrome, whereas the biomarkers IGFBP2 (insulin‐like growth factor‐binding protein 2) and NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide) were associated with a lower risk of incident metabolic syndrome.

What Are the Clinical Implications?

• There are many biological pathways that are shared between obesity and cardiometabolic disease.

• Examining proteomic profiles that are unique to cardiometabolic disease may offer insights into the mechanistic drivers of the transition from obesity to metabolic syndrome.

The prevalence of obesity has increased in recent decades, reaching >30% in the US population.¹ Obesity is associated with several metabolic abnormalities, including insulin resistance, hyperlipidemia, hypertension, and metabolic syndrome (MetS). Both obesity and MetS have been linked to increased cardiovascular disease (CVD) morbidity and mortality.², ³ Investigations into proteomic signatures of obesity and metabolic dysfunction have yielded important insights into the biological pathways that drive cardiometabolic disease. Previous work has implicated the insulin growth factor axis and inflammation in the regulation of obesity and insulin resistance.⁴, ⁵, ⁶

In this context, we sought to investigate proteomic signatures of cardiometabolic disease, including obesity, different adipose depots, and related clinical traits. We hypothesized that distinct aspects of cardiometabolic dysfunction would be associated with different characteristic proteins. We leveraged cardiovascular biomarkers ascertained as part of the National Heart, Lung, and Blood Institute’s Systems Approach to Biomarker Research in Cardiovascular Disease Initiative, which used high‐throughput technologies to identify novel circulating biomarkers in participants in the FHS (Framingham Heart Study).⁷ Using the Systems Approach to Biomarker Research in Cardiovascular Disease platform, we examined the associations of 71 protein biomarkers with cardiometabolic traits including traditional metabolic risk factors such as hypertension, dyslipidemia, and obesity as well as regional adiposity. We aimed to ascertain pathophysiological pathways that contribute to both obesity and metabolic disease, with the ultimate goal of identifying promising targets for the prevention and treatment of cardiometabolic disease.

Methods

Data Sharing

The data supporting the study findings will be made available on reasonable request. FHS data are made publicly available and can be accessed through the National Institutes of Health database of genotypes and phenotypes (https://www.ncbi.nlm.nih.gov/gap/).

Study Population

We examined proteomic profiles of participants in the Systems Approach to Biomarker Research in Cardiovascular Disease initiative of the FHS, a prospective longitudinal community‐based observational cohort study. Cross‐sectional analyses included all FHS Offspring cohort participants who attended examination 7 (1998–2001, n=3539) and FHS Generation 3 cohort participants who attended examination 1 (2002–2005, n=4095). Individuals with prevalent myocardial infarction (n=173), prevalent heart failure (n=24), end‐stage renal disease (n=24), missing covariates (n=288), and incomplete biomarker profiles (n=144) were excluded. The final cross‐sectional sample included 6981 individuals. For analyses involving computed tomography (CT) measures of adiposity, individuals with missing CT measures were further excluded (visceral adipose tissue [VAT]/subcutaneous adipose tissue [SAT], n=3840; pericardial/intrathoracic fat, n=3809; liver fat, n=3981). For longitudinal analyses, we included all FHS Offspring cohort participants who attended both examinations 7 and 8 and FHS Generation 3 cohort participants who attended both examinations 1 and 2, with a final sample of 4662 participants. Study protocols were approved by the appropriate institutional review boards, and all participants provided informed consent.

Clinical Assessment

Comprehensive medical histories, physical exams, and sample collections were obtained at each examination cycle. We defined obesity as body mass index (BMI) ≥30 kg/m². Hypertension was defined as systolic blood pressure ≥130 mm Hg, diastolic blood pressure ≥85 mm Hg, or current antihypertensive use. Diabetes mellitus (DM) was defined as fasting glucose (FG) ≥126 mg/dL, use of medication for DM, or reported history of DM. MetS was determined based on the National Cholesterol Education Program definition, which includes ≥3 of the following metabolic risk factors: abdominal obesity defined as waist circumference (WC) ≥88 cm for women and ≥102 cm for men, high triglyceride levels (≥150 mg/dL), low high‐density lipoprotein (HDL) levels (<50 mg/dL for women or <40 mg/dL for men), elevated blood pressure, high FG (≥100 mg/dL or current use of glucose‐lowering medications). Alcohol use was defined as ≥14 and ≥7 alcoholic drinks per week for men and women, respectively.

Intrathoracic, Pericardial, and Abdominal Fat Volume Measurements

Protocols for quantification of intrathoracic, pericardial, and abdominal fat have been described.⁸, ⁹ Quantification of all fat volumes involved measurement of fat‐containing pixels in a predefined image display setting based on Hounsfield units on a dedicated offline workstation (Aquarius 3D Workstation; TeraRecon Inc., San Mateo, CA). VAT and SAT were quantified by manually tracing the abdominal wall separating VAT and SAT. The interclass correlations of VAT and SAT were 0.992 and 0.997, respectively. Pericardial and intrathoracic fat volumes were determined using a semiautomatic segmentation technique. Pericardial fat volume was defined as adipose tissue within the pericardial sac. Total thoracic fat volume was defined as adipose tissue located within the thorax (from the level of the right pulmonary artery to the diaphragm and the chest wall to the descending aorta). Intrathoracic fat was derived from subtracting pericardial fat from total thoracic fat. The interclass correlation coefficients were 0.98 for total thoracic fat and 0.95 for pericardial fat. Liver fat content was quantified as the ratio of liver fat attenuation relative to a control (the liver phantom ratio). A total of 3 measures of at least 100 mm² in the liver, avoiding hepatic blood vessels, were performed. A calibration control (phantom) with CT‐Water and calcium hydroxyapatite was placed under each subject and used to standardize all liver measurements. The liver phantom ratio was calculated by dividing the mean of the 3 liver measures by the single phantom measure. Intrareader and interreader class correlation coefficients for the liver phantom ratios were 0.99 and 0.99, respectively.

Measurement of Circulating Biomarkers

Plasma concentrations of 85 candidate protein biomarkers were measured in the FHS Systems Approach to Biomarker Research in Cardiovascular Disease initiative. Candidate biomarkers were selected based on the association with atherosclerotic CVD, gene expression profiling, published genome‐wide associations studies of myocardial infarction and coronary heart disease, and discovery proteomics. Among the 85 biomarkers, 14 had few individuals above the lower limit of detection and were excluded, leaving 71 biomarkers for this analysis. Plasma proteins were quantified using a modified ELISA sandwich approach and Luminex xMAP platform (Sigma‐Aldrich, St. Louis, MO). Protocols for assay development have been previously presented. The assays demonstrated acceptable coefficients of variation from 2.0% to 9.5%. Assay characteristics are presented in Table S1.

Statistical Analysis

Baseline characteristics for the Offspring and Generation 3 cohort participants were summarized separately for the overall sample used in cross‐sectional analyses and for the sample included in the longitudinal analysis of incident MetS. Because of the skewed distributions of some proteins, values were rank normalized across all biomarkers. To assess the associations of single biomarker concentrations with metabolic traits, we performed linear regression analyses in an age‐adjusted and sex‐adjusted model and then a multivariable model additionally adjusting for systolic blood pressure, hypertension treatment, HDL, total cholesterol, DM, smoking, BMI, and log triglycerides. Liver fat analyses were also adjusted for alcohol use. Homeostatic Model Assessment of Insulin Resistance (HOMA‐IR) analyses further excluded participants with DM. To account for multiple testing in single biomarker models, a false discovery rate (FDR) q value <0.05 was set.¹⁰ For incident MetS, a logistic regression was performed where all prevalent cases were excluded from the analysis data set. Age and sex models were performed first, followed by a multivariable model additionally adjusting for systolic blood pressure, WC, hypertension treatment, FG, HDL, and log triglycerides. Statistical analyses for single biomarker analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

To better understand the association of cardiometabolic traits with potentially correlated protein biomarkers, we used hierarchical clustering and principal component analysis. Traits were grouped into discrete clusters based on biomarker profiles using hierarchical clustering (hclustvar function, ClustOfVar package, R). The optimal number of clusters was determined by maximizing average silhouette width (Fviz_nblust package in R). Next, we performed principal component analysis to reduce the dimensionality of the 71 biomarkers. We examined the associations of each of the 13 traits with the first 2 principal component terms (principal component 1 [PC1] and principal component 2 [PC2]). To visualize the associations of cardiometabolic traits with biomarker profiles, the β coefficients for each trait were projected onto the first 2 principal components (PC1 and PC2) to produce a vector plot. Inclusion of PC1 and PC2 explained up to 50% of the overall variance. Each trait vector demonstrates the directionality of each trait along the first 2 principal component terms (PC1 and PC2).

Results

A total of 6981 participants were included in the analysis (mean age 49 years; 54% women) with clinical characteristics displayed in Table 1. A total of 26% were obese (BMI ≥30 kg/m²), with a mean BMI of 27.4±5.5 kg/m² and WC of 96±15 cm. The prevalences of hypertension, DM, and MetS were 35%, 6%, and 33%, respectively. Mean visceral and subcutaneous adipose volumes were 1789±1015 cm³ and 2874±1388 cm³, with a mean VAT/SAT ratio of 0.68±0.39. Mean intrathoracic and pericardial fat volumes were 98±62 and 112±44 cm³. Over a median follow‐up of 6.3 years, 888 of 4662 (19%) participants free of MetS at baseline developed incident MetS.

Table 1.

Baseline Demographic and Clinical Characteristics

	Total	Offspring	Generation 3
	N=6981	N=3048	N=3933
Age, y	49 (14)	61 (9)	40 (9)
Women	3773 (54)	1678 (55)	2095 (53)
Body mass index, kg/m2	27.4±5.5	28.0±5.3	26.9±5.6
Waist circumference, cm	96±15	100±14	93±15
Fasting glucose, mg/dL	99±22	104±26	95±18
Systolic blood pressure, mm Hg	121±17	127±19	117±14
Diastolic blood pressure, mm Hg	75±10	74±10	75±10
Hypertension treatment	1303 (19)	982 (32)	321 (8)
Triglycerides, mg/dL	119 (109–130)	124 (114–137)	115 (107–124)
Total cholesterol, mg/dL	194±36	201±36	189±36
HDL cholesterol, mg/dL	54±17	54±17	54±16
Diabetes mellitus	390 (6)	316 (10)	74 (2)
Metabolic syndrome	2319 (33)	1444 (47)	875 (22)
HOMA‐IR, mg IU/dL·mL	6.1±5.5	4.0±3.5	7.7±6.2
Current smoker	997 (14)	396 (13)	601 (15)
CT measures of adiposity
Liver fat	0.36±0.05	0.35±0.05	0.36±0.05
VAT, cm3	1789±1015	2081±1086	1602±920
Intrathoracic fat, cm3	98±62	114±62	88±59
Pericardial fat, cm3	112±44	123±49	105±39
SAT, cm3	2874±1388	3013±1333	2785±1415
VAT/SAT ratio	0.68±0.39	0.76±0.44	0.63±0.34

Values are presented as number (percentage), mean±SD, or median (interquartile range). The following are the total sample sizes for CT measures of adiposity: liver fat, N=3000; VAT, N=3141; intrathoracic fat, N=3172; pericardial fat, N=3172; VAT/SAT ratio, N=3141. CT indicates computed tomography; HDL, high‐density lipoprotein; HOMA‐IR, Homeostatic Model Assessment of Insulin Resistance; SAT, subcutaneous adipose tissue; and VAT, visceral adipose tissue.

Multiple Biomarkers Associated With Obesity and Cardiometabolic Traits

Of the 71 biomarkers analyzed, 45 were significantly associated with obesity (multivariable‐adjusted FDR q value <0.05 for all listed in Table 2; full results are presented in Table S2). Among these 45 biomarkers, 32 were positively associated with obesity, with the largest associations observed for leptin, CRP (C‐reactive protein), and plasminogen activator inhibitor 1 (PAI1). A 1‐SD higher leptin level was associated with a near 7‐fold increased odds of obesity (odds ratio [OR], 6.99; 95% CI, 6.23–7.85; P=1.11×10⁻²³⁷). By contrast, 13 biomarkers were negatively associated with obesity, insulin‐like growth factor‐binding protein 1 (including IGFBP1) and insulin‐like growth factor‐binding protein 2 (IGFBP2). Higher levels of IGFBP1 and IGFBP2 were both associated with significantly lower odds of obesity (IGFBP1: OR, 0.37 [95% CI, 0.34–0.40], P=3.30×10⁻¹³⁴; IGFBP2: OR, 0.49 [95% CI, 0.45–0.52], P=2.55×10⁻⁸¹).

Table 2.

Protein Biomarkers With Significant Associations With Obesity

Biomarker	OR (95% CI)	P Value	Biomarker	OR (95% CI)	P Value
Positive Association			Negative Association
Leptin	6.99 (6.23–7.85)	1.11E‐237	IGFBP1	0.37 (0.34–0.40)	3.30E‐134
CRP	2.21 (2.05–2.37)	7.05E‐104	IGFBP2	0.49 (0.45–0.52)	2.55E‐81
PAI1	2.06 (1.91–2.21)	2.58E‐83	SRAGE	0.69 (0.64–0.73)	9.40E‐34
ADM	1.82 (1.69–1.96)	5.42E‐55	CNTN1	0.69 (0.65–0.74)	9.43E‐30
SAA1	1.81 (1.70–1.94)	2.67E‐68	CD56 or NCAM	0.69 (0.65–0.74)	1.05E‐27
ANGPTL3	1.77 (1.65–1.91)	2.30E‐56	OSTEO	0.73 (0.69–0.78)	6.84E‐23
UCMGP	1.69 (1.57–1.81)	1.51E‐46	IGF‐1	0.75 (0.71–0.80)	1.79E‐17
Adipsin	1.63 (1.52–1.75)	2.71E‐43	NT‐proBNP	0.82 (0.77–0.88)	1.61E‐08
AGP1	1.52 (1.42–1.62)	4.18E‐37	REG1A	0.82 (0.79–0.89)	1.10E‐09
FBN	1.49 (1.39–1.59)	7.48E‐33	BIKUNIN	0.84 (0.79–0.89)	1.71E‐08
EFEMP1	1.45 (1.35–1.56)	8.19E‐24	Leptin Receptor	0.89 (0.84–0.94)	0.0001
Hemopexin	1.41 (1.32–1.51)	3.65E‐25	SDF1	0.90 (0.85–0.95)	0.0003
C2	1.28 (1.20–1.36)	1.43E‐14	CD14	0.93 (0.88–0.99)	0.025
CXCL16	1.25 (1.18–1.33)	4.03E‐13
MPO	1.25 (1.18–1.33)	3.68E‐13
TIMP1	1.25 (1.16–1.34)	6.71E‐10
Cystatin C	1.24 (1.16–1.33)	1.50E‐09
BCHE	1.23 (1.15–1.31)	1.80E‐10
sICAM1	1.22 (1.15–1.30)	2.39E‐10
Resistin	1.22 (1.15–1.30)	8.22E‐11
MMP9	1.20 (1.13–1.28)	4.15E‐09
CD163	1.19 (1.12–1.26)	1.60E‐08
KLKB1	1.18 (1.11–1.26)	3.73E‐07
Ceruloplasmin	1.17 (1.10–1.26)	3.75E‐06
MCP1	1.17 (1.10–1.24)	1.02E‐06
B2M	1.16 (1.08–1.25)	2.16E‐05
MMP8	1.15 (1.08–1.22)	7.45E‐06
Myoglobin	1.13 (1.06–1.21)	0.0004
GMP140	1.11 (1.04–1.18)	0.0001
VEGF	1.09 (1.03–1.16)	0.003
CD5L	1.09 (1.02–1.15)	0.007
CD40L	1.08 (1.01–1.14)	0.02

Multivariable model adjusts for age, sex, smoking status, HDL, total cholesterol, hypertensiontreatment, log triglycerides, SBP, and diabetes. ADM, Adrenomedullin; AGP1, Alpha‐1 acid glycoprotein; ANGPTL3, Angiopoietin‐like 3; B2M, Beta‐2‐microglobulin; BCHE, Butyrylcholine esterase; BIKUNIN, AMBP‐bikunin; C2, Complement C2; CD14, Monocyte differentiation antigen; CD163, Cluster of differentiation 163; CD40L, Soluble CD40 ligand; CD56 or NCAM, Neural cell adhesion molecule; CD5L, CD5 antigen‐like; CNTN1, Contactin 1; CRP, C‐reactive protein; CXCL16, Chemokine (C‐X‐C motif) ligand 16; EFEMP1, EGF containing fibulin‐like extracellular matrix protein 1; FBN, Fibrinogen; GMP140, Granule membrane protein 140; HDL, indicated high‐density lipoprotein; IGF‐1, Insulin‐like growth factor 1; IGFBP1, Insulin‐like growth factor binding protein 1; IGFBP2, Insulin‐like growth factor binding protein 2; KLKB1, Plasma kallikrein; MCP1, Monocyte chemotactic molecule 1; MMP8, Matrix metallopeptidase 8; MMP9, Matrix metallopeptidase 9; MPO, Myeloperoxidase; NT‐proBNP, N‐terminal prohormone of brain natriuretic peptide; OR, odds ratio per 1‐SD increase in rank normalized biomarker; OSTEO, Osteocalcin; PAI1, Plasminogen activator inhibitor 1; REG1A, Lithostathine‐1‐alpha; SAA1, Serum amyloid A1; SBP, systolic blood pressure; SDF1, Stromal cell‐derived factor 1; sICAM1, Intercellular adhesion molecule 1; SRAGE, Receptor for advanced glycation endproducts; TIMP1, Tissue inhibitor of metalloproteinases 1; UCMGP, Uncarboxylated matrix Gla protein; VEGF, Vascular endothelial growth factor.

We next examined the associations of circulating biomarkers with 13 cardiometabolic traits including triglyceride level, insulin resistance as measured by HOMA‐IR, and CT measures of adiposity including SAT, VAT, pericardial, and intrathoracic fat depots as well as liver fat. We similarly found that numerous proteins were significantly associated with each of these cardiometabolic traits (Figure 1; full results are presented in Table S3). Of 71 biomarkers, 68 were significantly associated with 1 or more of the 13 examined cardiometabolic traits in multivariable adjusted analyses (FDR q value <0.05 for all). We observed the greatest number of biomarker associations with triglyceride levels. Specifically, 53 of the 71 biomarkers were significantly associated with triglyceride levels (FDR q value <0.05 for all; P value range 1.40×10⁻¹¹⁷⁹ to 0.024). Among those 53 biomarkers, 38 were positively associated with triglycerides, including apoA1 (apolipoprotein A1) and KLKB1, whereas 15 were negatively associated with triglycerides, including IGFBP2 and tetranectin. A 1‐SD higher apoA1 level was associated with 0.19 SD higher log triglyceride levels (ß, 0.19; SE, 0.01; P=1.48×10⁻⁶³), whereas a 1‐SD higher IGFBP2 level was associated with 0.23‐SD lower log triglyceride levels (β, −0.23; SE, 0.01; P=1.40×10⁻¹¹⁹).

Figure 1. Proteomic signatures of obesity and related cardiometabolic traits.

Heatmap of associations of single biomarkers with cardiometabolic traits. Color coding represents standardized β coefficient in multivariable‐adjusted analyses (X‐SD change in trait per 1‐SD change in biomarker). Clustering based on biomarker correlations (rows) and traits (columns). BMI indicates body mass index; DBP, diastolic blood pressure; FG, fasting glucose; HDL, high‐density lipoprotein; HOMA‐IR, Homeostatic Model Assessment of Insulin Resistance; SAT, subcutaneous adipose tissue; SBP, systolic blood pressure; trig, triglycerides; VAT, visceral adipose tissue; and WC, waist circumference. For abbreviations of biomarkers, please see Table S1.

Protein biomarkers significantly associated with CT measures of regional adiposity, including SAT, VAT, pericardial, intrathoracic, and liver fat (Figure 1). In multivariable‐adjusted analyses, 22 biomarkers were associated with VAT, 16 with SAT, 15 with pericardial fat, 17 with intrathoracic fat, and 26 with liver fat (FDR q value <0.05 for all; P value range 0.01 to 1.82×10⁻⁴⁰). Of the 22 biomarkers significantly associated with VAT, higher concentrations of 8 biomarkers, including PAI1, AGP1 (alpha‐1 acid glycoprotein), and leptin, were associated with higher VAT levels, whereas higher levels of the remaining 14 biomarkers were associated with lower levels of VAT, including IGFBP1, IGFBP2, and CNTN1 (contactin 1). A 1‐SD higher PAI1 level was associated with a 0.11 unit higher VAT (β, 0.11; SE, 0.01; P value 2.21×10⁻¹⁸), whereas a 1‐SD higher IGFBP2 level was associated with a 0.13 unit lower VAT (β, −0.13; SE, 0.01; P value 7.57×10⁻²⁴).

Overlapping Versus Distinct Proteins of Obesity and Cardiometabolic Disease

We observed significant overlaps of biomarker profiles of obesity, cardiometabolic clinical traits, and measures of regional adiposity (Figure 2). When superimposing biomarker profiles of obesity and metabolic traits (triglycerides, FG, liver fat, and HOMA‐IR), we found that all 45 biomarkers associated with obesity were also associated with 1 or more metabolic traits and 7 biomarkers (butyrylcholine esterase (BCHE), intercellular adhesion molecule 1 (sICAM1), PAI1, IGFBP1, granule membrane protein 140 (GMP140), B‐type natriuretic peptide (NT‐proBNP), IGFBP2) were shared by all 4 traits (Figure 2A). We also examined the overlap of obesity with CT measures of adiposity (Figure 2B). Specifically, 4 biomarkers (leptin, PAI1, AGP1, CNTN1) were common to obesity and all measures of regional adiposity. VAT, intrathoracic fat, and pericardial fat shared similar proteomic signatures. Of the 22 biomarkers associated with VAT, 13 were also associated with pericardial fat and 10 were associated with intrathoracic fat. Although SAT also shared similar proteomic profiles with VAT, intrathoracic fat, and pericardial fat, 6 biomarkers (ceruloplasmin, complement C2 (C2), Cystatin C, FBN, adipsin, and adrenomedullin (ADM)) were uniquely associated with obesity and SAT. Results were similar when restricting the analyses to participants with available CT measures of adiposity only (Table S4 and Figure S1).

Figure 2. Overlap of biomarker profiles of obesity, cardiometabolic clinical traits, and computed tomography measures of regional adiposity.

A, Overlap between biomarkers associated with obesity and cardiometabolic traits. B, Overlap between biomarkers associated with obesity and computed tomography measures of adiposity. FG indicates fasting glucose; HOMA‐IR, Homeostatic Model Assessment of Insulin Resistance; SAT, subcutaneous adipose tissue; TG, triglycerides; and VAT, visceral adipose tissue.

Biomarker Profiles Identify Cardiometabolic Trait Clusters

Despite significant overlap in some proteins across cardiometabolic traits, we recognized that distinct traits also displayed differences. We sought to identify clusters of traits based on proteomic signatures. In hierarchical clustering analyses, we identified 3 unique cardiometabolic trait clusters with distinct biomarker profiles: (1) adipose traits, (2) metabolic traits, and (3) lipid traits (Figure 1). Adipose traits included BMI, WC, SAT, VAT, intrathoracic fat, and pericardial fat. Metabolic traits included triglycerides, HOMA‐IR, liver fat, FG, diastolic blood pressure, and systolic blood pressure, and lipid traits included HDL. In a complementary analysis using principal component analysis, we similarly found that distinct cardiometabolic traits displayed different proteomic architecture. Based on the first and second principal component terms, traits separated along the following vector directions based on biomarker profiles (Figure 3): BMI and WC separated along the same vector direction, and SAT followed a parallel but opposite vector direction, suggesting close alignment in biomarker profiles for these 3 traits. HDL and triglycerides separated along a vector that was orthogonal to BMI, WC, and SAT, which suggests that the biomarker profiles of lipid traits (HDL and triglycerides) were uniquely different from the proteomic signature of fat traits (BMI, WC, and SAT).

Figure 3. Vector map of PC1 and PC2 calculated using principal components analysis.

BMI indicates body mass index; DBP, diastolic blood pressure; FG, fasting glucose; HDL, high‐density lipoprotein; PC1, principal component 1; PC2, principal component 2; SAT, subcutaneous adipose tissue; SBP, systolic blood pressure; Trig, triglycerides; VAT, visceral adipose tissue; and WC, waist circumference.

Biomarkers Predict Incident MetS

Finally, we evaluated the association of protein biomarkers with incident MetS in age‐and sex‐adjusted models and a multivariable‐adjusted model (Figure 4; full results are presented in Table S5 through S6). Of 71 biomarkers, 30 were significantly associated with incident MetS in age‐adjusted and sex‐adjusted models (FDR q value <0.05). Among these 30 biomarkers, 21 biomarkers were associated with an increased odds of developing MetS, including apoB (apolipoprotein B), leptin, and CRP, whereas 9 were associated with a decreased odds of incident MetS, including IGFBP1, IGFBP2, and NT‐proBNP. Of the 30 biomarkers, 6 remained significant after multivariable adjustment, including apoB, IGFBP2, KLKB1, C2, FBN, and NT‐proBNP. IGFBP2 and NT‐proBNP were associated with a decreased odds of incident MetS. A 1‐SD higher IGFBP2 concentration was associated with a 0.83 decreased odds of future MetS (OR, 0.83; 95% CI, 0.75–0.92; P value 0.0005). Conversely, apoB, KLKB1, C2, and FBN were associated with an increased odds of future MetS; a 1‐SD higher apoB concentration was associated with a 1.50 increased odds of incident MetS (OR, 1.50; 95% CI, 1.36–1.66; P value 4.97×10⁻¹³).

Figure 4. Association of single biomarkers with incident metabolic syndrome. Biomarkers shown were statistically significant in age‐adjusted and sex‐adjusted models (false discovery rate q value <0.05).

*Biomarkers that were statistically significant in multivariable adjusted models (false discovery rate q value <0.05). ORs displayed per 1‐SD increase in biomarker concentration. Black represents the age‐adjusted and sex‐adjusted model. Red represents the multivariable model. Multivariable model adjusts for age, sex, waist circumference, high‐density lipoprotein, log triglycerides, systolic blood pressure, hypertension treatment, fasting glucose, and statin therapy. OR indicates odds ratio. For biomarker abbreviations, please see Table S1.

Discussion

We examined the proteomic profiles of obesity and associated cardiometabolic traits, including specific adipose depots in a community‐based sample of adults free of CVD. Our findings are 4‐fold: (1) numerous CVD‐related protein biomarkers are associated with obesity, cardiometabolic clinical traits, and CT measures of regional adiposity; (2) the proteomic architecture of obesity overlaps considerably with associated cardiometabolic traits, implying shared pathways; (3) despite overlap, hierarchical clustering of proteomic profiles identified 3 distinct clusters of cardiometabolic traits (adipose, metabolic, and lipid traits); and (4) 6 protein biomarkers were associated with a future risk of MetS. Collectively, our findings highlight new associations of CVD biomarkers with obesity, cardiometabolic traits, and CT measures of adiposity and suggest that proteomic profiles may refine our phenotypic characterization of obesity and cardiometabolic disease by elucidating both unique and shared biological pathways.

Although the link between obesity and cardiometabolic disease is well established, biological pathways involved in the pathogenesis of obesity‐related diseases are complex and incompletely understood. Previously established biomarkers for obesity are implicated in pathways of inflammation, adipogenesis, and cell proliferation.⁴, ¹¹ Small‐scale proteomic‐targeted approaches have sought to better understand the mechanistic pathways that link obesity with cardiometablic disease. A study of 56 healthy middle‐aged overweight subjects examined the association of 124 plasma proteins with BMI and plasma insulin.⁶ The authors identified 3 clusters of plasma proteins associated with BMI and 4 protein clusters associated with insulin. Proteins strongly associated with both BMI and insulin included complement 3, CRP, serum amyloid protein, and VEGF (vascular endothelial growth factor), highlighting the importance of inflammation in the pathogenesis of both obesity and DM. We extend these findings and demonstrate novel biomarkers shared among obesity and cardiometabolic traits. Our observations corroborate existing data: higher levels of adipokines leptin, adiponectin, adipsin, and CRP, and lower levels of IGF‐1 (insulin‐like growth factor 1), IGFBP1, IGFBP2, and insulin‐like growth factor‐binding protein 3 (IGFBP3) were significantly associated with obesity and several cardiometabolic traits. We also show that other markers of inflammation, fibrosis, and angiopoietins were among the protein biomarkers with the strongest associations with obesity. For instance, higher levels of PAI1, a physiologic inhibitor of plasminogen activators, and ANGPTL3, an angiopoietin involved in angiogenesis and the regulation of circulating HDL and triglyceride levels, were both significantly associated with obesity and numerous cardiometabolic traits.

An important novelty of our study is the examination of protein biomarker associations with regional adiposity measures. Although traditional metrics of obesity (BMI and WC) are well correlated with cardiometabolic risk, they do not accurately capture the differential contributions of different fat compartments to metabolic risk. Previous work has demonstrated that VAT, SAT, pericardial fat, and intrathoracic fat are all variably associated with metabolic risk, with VAT being the most strongly predictive of cardiometabolic risk and CVD.⁸, ¹², ¹³, ¹⁴ In a study of 1583 participants from the FHS, biomarkers secreted by adipose tissue (adiponectin, leptin, leptin receptor, and fatty acid‐binding protein 4) and those secreted by both adipose tissue and the liver (fetuin‐A and retinol binding protein 4) were significantly associated with SAT and VAT.¹⁵ In our analysis, 4 biomarkers were significantly associated with all 4 measures of adiposity and obesity: leptin, PAI1, AGP1 (positively associated), and CNTN1 (negatively associated). Although the associations of leptin and PAI‐1 with obesity have been described,¹⁶, ¹⁷ little is known about the associations of AGP1 and CNTN1 with cardiometabolic disease. In obese mice, AGP1 expression was induced by metabolic and inflammatory signals in adipose tissue and was protective against the deleterious effects of severe inflammation.¹⁸ In a study of 64 obese individuals undergoing weight reduction surgery, circulating levels of AGP1 were associated with BMI, leptin, fasting insulin, HOMA‐IR, and CRP. There were no differences in protein AGP1 expression between VAT and SAT.¹⁹ CNTN1 is a neuronal membrane protein required for axonal growth and maturation. In one study examining regional differences in subcutaneous adipose tissue gene expression, CNTN1 was differentially expressed in the lower abdomen compared with the flank.²⁰ Interestingly, the biomarker profiles of VAT, pericardial fat, and intrathoracic fat were closely aligned, whereas SAT appeared to carry a unique proteomic profile. Emerging data suggest that visceral adiposity, rather than subcutaneous fat, is associated with increased metabolic risk.¹⁵, ²¹ Of note, the insulin growth factor axis demonstrated strong inverse correlations with VAT, pericardial and intrathoracic fat, but not with SAT. The insulin growth factor axis is an important regulator of obesity and metabolic risk; its lack of association with SAT may explain in part why SAT only conveys minimal cardiometabolic risk compared with other fat compartments.

Although we observed many shared protein biomarker associations between obesity and multiple cardiometabolic traits, there were additional biomarkers that were uniquely associated with 1 or more cardiometabolic clinical traits. For example, melanoma cell adhesion molecule (MCAM), growth differentiation factor 15 (GDF15), and lipoprotein(a) (LPA) were significantly associated with FG, HOMA‐IR, triglycerides, and liver fat, but not with obesity. The notion that distinct proteomic profiles exist was further corroborated by clustering analyses and principal component analysis, which identified 3 distinct clusters of cardiometabolic traits: adipose, metabolic, and lipid. These findings suggest that in addition to shared biology, there may be pathways that are related to metabolic dysfunction independent of generalized adiposity. These may represent a precursor phenotype to obesity and its related diseases. This assertion is supported by the concept of obesity without MetS or the metabolically healthy obesity phenotype.²² In a study of 6809 participants from the Multi‐Ethnic Study of Atherosclerosis, investigators found that the risk of incident CVD was not increased in participants with the metabolically healthy obesity phenotype, although nearly 50% of individuals with the metabolically healthy obesity phenotype at baseline eventually developed MetS with a subsequent increased risk of CVD.²³ Ascertaining which biologic pathways are activated in cardiometabolic disease may refine our understanding of the mechanistic drivers of the transition from metabolically healthy obesity to MetS. For instance, GDF15 and lipoprotein(a) have both been associated with cardiometabolic risk, and further exploration of their associated pathways may offer insights into that transition.²⁴, ²⁵ Moreover, duration of obesity may influence proteomic profiles as well as the development of cardiometabolic disease. Future longitudinal investigations examining the association of time‐dependent changes in biomarkers with cardiometabolic traits in obesity may also provide biomarker targets for further study and potential intervention.

Finally, we identified 6 unique protein biomarkers with significant associations with incident MetS in multivariable analyses. Higher levels of IGFBP2 and NT‐proBNP were associated with a lower risk of future MetS, whereas increased levels of apoB, KLKB1, C2, and FBN were associated with a higher risk of future MetS. IGFBP2 levels are consistently lower among patients with MetS, and lower levels are associated with unfavorable metabolic risk factors such as higher BMI, lower insulin sensitivity, and less favorable lipid profiles.²⁶, ²⁷, ²⁸ Experimental evidence suggests that after fetal development, IGFBP2 inhibits adipogenesis, improves insulin sensitivity, and significantly attenuates the risk of developing DM.²⁹, ³⁰, ³¹ NT‐proBNP was also protective against risk of MetS, even after adjustment for BMI. It is well known that natriuretic peptide levels are inversely related to BMI and obesity.³², ³³, ³⁴ Given the strong negative association of natriuretic peptides with obesity, the relationship between NT‐proBNP and MetS is anticipated, but whether this association persists independent of obesity is not known. Among the biomarkers associated with increased risk of MetS, the relationship of apoB with incident MetS has been the most well described, although the driving mechanisms are still being elucidated.³⁵, ³⁶, ³⁷ In 10 340 Chinese participants with MetS, serum apoB was associated with an increased risk of prevalent MetS, particularly among individuals with a healthy weight.³⁷ This observation suggests that apoB may exert its effects on MetS risk via both obesity‐dependent and obesity‐independent pathways. apoB is intimately linked to lipid metabolism, and we found that many biomarkers were shared between metabolic and lipid traits, but not obesity or fat traits, suggesting that lipid pathways may contribute to the development of metabolic disease independent of obesity. Less is known about the relationship of KLKB1, C2, and FBN with MetS, but these findings point to the involvement of coagulation, complement activation, and fibrosis pathways in the pathogenesis of MetS.

Our study has several limitations. First, the selection of biomarkers was based on previously established associations with CVD. Given the significant overlap between metabolic disease and established CVD, our investigation was enriched for protein biomarkers with significant associations with cardiometabolic traits. A more unbiased selection strategy may have yielded additional novel targets for future discovery. Second, our study population included predominantly White participants, limiting the generalizability to other diverse populations. Metabolic disease is more prevalent among non‐White populations, especially Black women, and protein targets may dramatically differ in other populations.³⁸ Finally, our study is an observational investigation, and conclusions about causation cannot be drawn.

We demonstrate significant associations of CVD‐related protein biomarkers with obesity, cardiometabolic clinical traits, and CT measures of regional adiposity. Protein biomarkers clustered along 3 main axes: adipose, metabolic, and lipid traits. Although obesity shared similar proteomic profiles with fat, metabolic, and lipid traits, several biomarkers were unique to metabolic traits only. Further exploration of these novel protein targets and associated pathways may provide insight into the mechanisms responsible for the progression from obesity to cardiometabolic disease.

Sources of Funding

This work was supported by grants from the National Institutes of Health (5T32HL094301‐07 to E.S. Lau), National Heart, Lung, and Blood Institute (1K22Hl135075‐01 to J. Ma), National Heart, Lung, and Blood Institute Division of Intramural Research (to D. Levy), and National Institutes of Health contracts N01‐HC‐25195 and HHSN2682015000011 (Framingham Heart Study) and R01‐HL134893, R01‐HL140224, and K24 HL 153669 (to J.E. Ho).

Disclosures

Dr Ho has received research support from Gilead Sciences and Bayer AG and research supplies from EcoNugenics. The remaining authors have no disclosures to report.

Supporting information

Table S1–S6

Figure S1

(J Am Heart Assoc. 2021;10:e020215. DOI: 10.1161/JAHA.120.020215.)