Metabolomic Profile of the Healthy Eating Index-2015 in the Multiethnic Study of Atherosclerosis

Abstract

Background

Poor diet quality is a risk factor for type 2 diabetes and cardiovascular disease. However, knowledge of metabolites marking adherence to Dietary Guidelines for Americans (2015 version) are limited.

Objectives

The goal was to determine a pattern of metabolites associated with the Healthy Eating Index (HEI)-2015, which measures adherence to the Dietary Guidelines for Americans.

Methods

The analysis examined 3557 adult men and women from the longitudinal cohort Multiethnic Study of Atherosclerosis (MESA), without known cardiovascular disease and with complete dietary data. Fasting serum specimens and diet and demographic questionnaires were assessed at baseline. Untargeted ¹H 1-dimensional nuclei magnetic resonance spectroscopy (600 MHz) was used to generate metabolomics and lipidomics. A metabolome-wide association study specified each spectral feature as outcomes, HEI-2015 score as predictor, adjusting for age, sex, race, and study site in linear regression analyses. Subsequently, hierarchical clustering defined the discrete groups of correlated nuclei magnetic resonance features associated with named metabolites, and the linear regression analysis assessed for associations with HEI-2015 total and component scores.

Results

The sample included 50% women with an mean age of 63 years, with 40% identifying as White, 23% as Black, 24% as Hispanic, and 13% as Chinese American. The mean HEI-2015 score was 66. The metabolome-wide association study identified 179 spectral features significantly associated with HEI-2015 score. The cluster analysis identified 7 clusters representing 4 metabolites; HEI-2015 score was significantly associated with all. HEI-2015 score was associated with proline betaine [β = 0.12 (SE = 0.02); P = 4.70 × 10⁻¹³] and was inversely related to proline [β = −0.13 (SE = 0.02); P = 4.45 × 10⁻¹⁴], 1,5 anhydrosorbitol [β = −0.08 (SE = 0.02); P = 4.37 × 10⁻⁷] and unsaturated fatty acyl chains [β = 0.08 (SE = 0.02); P = 8.98 × 10⁻⁷]. Intake of total fruit, whole grains, and seafood and plant proteins was associated with proline betaine.

Conclusions

Diet quality is significantly associated with unsaturated fatty acyl chains, proline betaine, and proline. Further analysis may clarify the link between diet quality, metabolites, and pathogenesis of cardiometabolic disease.

Introduction

Poor diet quality is independently associated with the incidence of CVD [1,2], cancer [3], and type 2 diabetes (T2D) [[4], [5], [6]]. The Healthy Eating Index (HEI)-2015 is a measure of diet quality reflecting adherence to the Dietary Guidelines for Americans (DGA) 2015–2020 [7]. The DGA 2015–2020 represents dietary guidance jointly published by the United States Department of Agriculture and the United States Department of Health and Human Services every 5 years, reflecting recommendations for ideal intake, by the United States Government. An important update to the HEI-2015 from earlier versions is a recommendation to limit intakes of both added sugars and saturated fats to <10% of energy.

The identification of small molecules, called metabolites, present in serum, urine, or tissue, may help to shed light on the phenotypic links between habitual diet quality and disease. Diet quality is a complex, long-term exposure, likely affecting multiple metabolic processes simultaneously, and habitual diet intake may produce a stable metabolic environment that is linked with risk of diseases. Previous assessments of the HEI-2015 score and associated metabolites have been limited to targeted or commonly annotated metabolites, which may not capture the full metabolome representing consumption of a higher quality diet [8]. Previous work has also demonstrated that there may be stronger links between diet-associated circulating metabolites and diseases than the original association between diet quality and disease outcomes [[9], [10], [11]]. A deeper assessment using NMR-based spectral features may allow for a more nuanced assessment of diet quality, which may support future assessment of diet quality and association with disease.

The objective of this investigation was to determine a pattern of metabolites associated with habitual diet quality as represented by the HEI-2015 and its components. This analysis profiled serum untargeted NMR-based metabolomics to gain insight into metabolic features associated with high-diet quality.

Methods

Participants

We included 3557 adult men and women, determined through self-reported sex, from the Multiethnic Study of Atherosclerosis (MESA) longitudinal cohort study without known CVD at enrollment visit and with stored serum samples with NMR-based Combinatorial Biomarkers for subclinical atherosclerosis [12] metabolomic profiling data available for the analysis. MESA is a United States–based prospective cohort study of 6814 participants between the ages of 45 and 84 years recruited at 6 sites (Baltimore City and County, MD; Chicago, IL; Forsyth County, NC; New York, NY; Los Angeles County, CA; and St. Paul, MN), designed to investigate the development and progression of subclinical atherosclerotic disease. Participants were enrolled between 2000 and 2002 [13] did not experience CVD at baseline and were purposively recruited from 4 race/ethnicity categories (Black, White, Chinese American, and Hispanic). An institutional review board approval was obtained at all participating centers, and all participants gave informed consent.

We included 3663 participants with available metabolomics data from the baseline examination. We further excluded 106 with implausible caloric intake (<600 or >6000 kcal/d, in concordance with previous MESA publications [2,4,14]) or who were missing two-thirds or more of diet data. Of these, 3557 participants had available metabolomics measures from the baseline examination.

Data and biospecimens

We assessed clinical and demographic data using questionnaires administered at baseline. Fasting biospecimens were collected at baseline and stored at −80 °C until analyzed. Participants were asked to fast for 12 h, avoid smoking on the morning of the examination, and avoid heavy exercise 12 h before the examination.

Metabolomic profiling

NMR measurements were performed according to a previously published protocol using serum samples [15]. In brief, a standard ¹H NMR 1-dimensional (1D NMR) spectrum with water suppression was obtained for each sample, detecting signatures of all proton containing compounds, such as sharp peaks from small molecule species and broad peaks from lipoproteins and proteins. Subsequent spectral processing was performed using the software TOPSPIN 3.1 (Bruker Biospin). The spectra were automatically phased and baseline corrected, and the chemical shifts were calibrated to the glucose signal at 5.233 ppm. Spectral data were imported into MATLAB (version 8.3 [R2014a]; Mathworks Inc.) for further processing, such as peak alignment and normalization using PQN method [16].

The spectral features were annotated using the following spectral information: chemical shift (ppm), the coupling constant (J in hertz), the peak multiplicity (singlet, doublet, and multiplet), and peak connectivity of the NMR signals from the 1D and 2D NMR spectra (2D JRES, correlation spectroscopy, total correlation spectroscopy, and heteronuclear single quantum correlation spectroscopy) and statistical correlation methods (statistical total correlation spectroscopy and subset optimization by reference matching) [17]. Annotations were also assessed using information from available in-house and publicly available spectral databases and with published data.

Diet assessment

Usual dietary intake over the past 12 mo was assessed at baseline, from a self-administered 120-item FFQ, which evaluated diet intake over the past year. The MESA FFQ is a modified version of the Insulin Resistance Atherosclerosis Study FFQ, which was previously validated in those identified as non-Hispanic Whites, Hispanic ethnicity, and Black [18]. The MESA FFQ was modified from that used in Insulin Resistance Atherosclerosis Study to include dietary intake common among Chinese Americans. For each food item, participants indicated the typical serving size and the frequency each food was eaten. Frequency ranged from “rare or never” to a maximum of “2+ times per day” for foods and “6+ times per day” for beverages. Mean daily servings of 47 food groups were created using weighted recipes from the Nutrition Data System for Research and estimated per 100 g of food and were used as the basis for creation of the diet score.

HEI-2015 score

The HEI-2015 was designed to align with the DGA 2015–2020 [7]. The HEI-2015 contains 13 components, the sum of which totals to a maximum score of 100 points. As in HEI-2005 and HEI-2010, each of the components is scored on a density basis of 1000 calories, with the exception of FAs, which is a ratio of unsaturated to SFAs.

There are 9 adequacy components: total fruits, whole fruits, total vegetables, greens and beans, whole grains, dairy, total protein foods, seafood and plant proteins, and FAs for which greater consumption is the goal. For 4 moderation components, we assigned higher scores with minimization of intake for the following food groups: refined grains, sodium, added sugars, and saturated fats.

Statistical analysis

Metabolome-wide association study

The association of all 30,590 spectral features, which were mean-centered and scaled to unit variance, with HEI-2015 score was run using linear regression models specifying each spectral feature as the outcome in separate models, with SD of HEI-2015 score as the predictor, and age (continuous), race (categorical), sex (binary), and data collection site (categorical) as covariates. A spectral decomposition based on the correlation matrix between all spectra suggested that the effective number of independent tests was 22,857. Therefore, significance for associations between spectral features and HEI-2015 score was set at Bonferroni-corrected significance level of P < 2.2 × 10⁻⁶ (0.05/22,857).

Elastic-net–regularized regression

To adjust for unreliable parameter estimates that may occur when using multiple regression models in the setting of multicollinearity, we performed an elastic-net–regularized regression model to evaluate metabolites that were significant in independent analyses. The elastic-net model allowed for a penalized logistic regression on all biomarkers simultaneously to identify the metabolites most highly associated with diet pattern score. Elastic-net–regularized regression models were run with HEI-2015 diet score as the predictor and spectral features showing a significant association with the HEI-2015 diet score in metabolome-wide association study (MWAS) analysis as the outcomes. Optimal penalty parameters for the penalty value (mixing percentage, α) and the strength of the penalty (regularization penalty, λ) were ascertained through the package caret in R using cross-validation. In briefly, data in the full data set were randomly assigned to one of the 2 equal-sized data sets. Parameter selection was performed using resampling of models with 100 values of λ chosen according to the caret algorithm. Then, the final selected parameters were applied to analyses on the whole data set. Optimization was reached using feature-wise normalization change in successive coordinate descent iterations. Model performance was judged based on root mean square error of approximation, with α and λ parameters giving rise to the minimum mean cross-validated error used to generate new coefficients for the association of spectral features with HEI-2015 score.

Clustering analysis

Pearson correlations were run between all spectral features with nonzero coefficients in the elastic-net–regularized regression models, to allow for identification of clusters or groups of spectral features. Because groups of spectral features showed specific patterns of intercorrelations, all spectral features with nonzero coefficients from the regularized regression models were subject to a hierarchical clustering analysis. The hierarchical clustering analysis was performed using the package NbClust in R. Euclidean distance was used to compute the dissimilarity matrix, with total within-cluster variance computed using Ward (1963) algorithm to minimize the total within-cluster variance. The optimal number of clusters was identified using the Duda-Hart stopping rule. For clusters with contributions of spectral features from >1 annotation, we assigned the metabolite with the most prominent signals. Methanol or proline was assigned as proline owing to the presence of a coefficient of association of proline with the same β coefficient as proline or methanol, and histidine or proline betaine was assigned as proline betaine based on the absence of nonoverlapping signals from histidine within that spectral feature.

Final associations between HEI-2015 diet score and metabolomics cluster scores

Because several spectral features may be representative of the same metabolite, to assist in interpretability and most accurately represent the presence of individual metabolites, sum scores for all spectral features within a cluster were created. Based on the annotations assigned to the spectral feature, the most likely metabolite or metabolites represented by each cluster score was assigned. Cluster scores were highly skewed, thus were winsorized and represented as 4 SDs ± the mean and transformed using a blom transformation. Associations were analyzed from linear regression models with HEI-2015 diet score standardized using z-score as the predictor, transformed cluster scores as the outcomes in separate models, and age, sex, race and site of data collection as fixed effects and were standardized. Significance was retained as a Bonferroni correction for the original number of effective number of independent tests in the MWAS (of P < 2.2 × 10⁻⁶). For all cluster scores significantly associated with HEI-2015, multivariable linear regression models were run with the cluster score the outcomes in separate models, all 13 components of HEI-2015 score as the predictors within the same model, and age, sex, race, and site of data collection as fixed effects. Significance was set at a Bonferroni correction for 7 tests (0.05/7 = P < 0.007).

Results

The sample of participants self-identified as 50% women and 13% of participants as Black, 23% of Hispanic ethnicity, 24% Chinese American and 40% non-Hispanic White, with a mean age of 63 years (Table 1). The mean HEI-2015 score was 66. The HEI-2015 score was significantly associated with 179 1D NMR-based spectral features determined through the MWAS analysis. (Supplemental Table 1 and Supplemental Figure 1).

TABLE 1.

Baseline characteristics of MESA cohort participants by Healthy Eating Index (HEI)-2015 quartile (N = 3557)

	Mean (SD)	Quartile 1	Quartile 2	Quartile 3	Quartile 4
N	3557	847	880	907	923
Women, n (%)	1787 (50)	334	403	465	585
Age, y	63 (10)	60 (10)	62 (10)	63 (10)	65 (10)
Race, n (%)
White	1428 (40)	337 (40)	318 (36)	368 (41)	405 (44)
Black	830 (23)	229 (27)	208 (24)	178 (20)	215 (23)
Hispanic	838 (24)	148 (17)	221 (25)	260 (29)	209 (23)
Chinese American	461 (13)	133 (16)	133 (15)	101 (11)	94 (10)
HEI-2015 score	66 (8)	56 (4)	64 (2)	69 (1)	76 (4)
BMI (kg/m2)	28 (5)	28 (6)	29 (6)	28 (5)	28 (5)
Diabetes, n (%)	470 (13)	99 (14)	114 (13)	136 (15)	121 (13)
Hypertension, n (%)	1608 (45)	365 (43)	387 (44)	407 (45)	449 (49)

HEI, Health Eating Index.

The clustering analysis identified 7 main clusters of metabolomic spectral features, each identified by a single metabolite or lipid (Table 2 and Supplemental Figure 2). Four of the 7 clusters contained spectral features annotated to the amino acid proline. A higher HEI-2015 score, reflecting a better diet quality, was associated with a lower abundance of proline (P < 0.007, corrected for 7 cluster comparisons). The strongest association was found between HEI-2015 score and proline betaine [β = 0.12 (SE = 0.02); P = 4.70 × 10⁻¹³].

TABLE 2.

Associations of Healthy Eating Index 2015 diet score with representative metabolites and lipids

Cluster	Spectral features	Metabolite association	β1	SE	P
1	2.765603, 2.769304, 2.769641, 2.770313, 2.77065	C=CHCH2HC=C (fatty acyl chains)	0.08	0.02	8.98 × 10−7
2	3.100354, 3.10069, 3.101027, 3.101363	Proline betaine/histidine	0.12	0.02	4.70 × 10−13
3	3.268907	1,5-Anhydrosorbitol	−0.08	0.02	4.37 × 10−7
4	3.3261, 3.326437	Proline	−0.09	0.02	5.46 × 10−8
5	3.342249, 3.347968	Methanol/proline	−0.10	0.02	4.06 × 10−10
6	3.34494, 3.345277	Methanol/proline	−0.12	0.02	1.63 × 10−12
7	3.34595, 3.346286	Methanol/proline	−0.13	0.02	4.45 × 10−14

¹Standardized estimates: adjusted for age, sex, race, and study sites.

Intake of specific HEI-2015 components was differentially associated with the defined clusters of metabolomic spectral features (Table 3). The HEI-2015 score component “total dairy” was associated with 4 clusters, representing 1,5-anhydrosorbitol and methanol or proline. Higher intake of dairy products was linked with lower abundance of both metabolites, mirroring the findings of total HEI-2015 score and these metabolites.

TABLE 3.

Associations of Healthy Eating Index 2015 component scores with metabolomic cluster scores

Component	Cluster	Most likely annotation	β1	SE	P
Total fruit	2	Proline betaine/histidine	0.18	0.02	3.25 × 10−12
Whole grains	2	Proline betaine/histidine	0.05	0.02	2.54 × 10−3
Total dairy	3	1,5-Anhydrosorbitol	−0.06	0.02	1.29 × 10−3
	5	Methanol/proline	−0.12	0.01	7.31 × 10−10
	6	Methanol/proline	−0.11	0.01	2.86 × 10−9
	7	Methanol/proline	−0.13	0.01	1.28 × 10−11
Total protein	5	Methanol/proline	−0.10	0.03	1.74 × 10−4
	6	Methanol/proline	−0.11	0.03	3.65 × 10−5
	7	Methanol/proline	−0.11	0.03	2.42 × 10−5
Seafood and plant protein	2	Proline betaine/histidine	0.08	0.02	1.05 × 10−3
FA	1	C=CHCH2HC=C (fatty acyl chains)	0.02	0.01	3.83 × 10−3
Refined grains	5	Methanol/proline	−0.07	0.02	7.07 × 10−5
	6	Methanol/proline	−0.06	0.02	3.12 × 10−4
	7	Methanol/proline	−0.06	0.02	2.52 × 10−4

¹Standardized estimates: adjusted for age, sex, race, and study sites.

Intake of the HEI-2015 component “total fruits” had strong, positive associations with proline betaine [β = 0.18 (SE = 0.02); P = 3.24 × 10⁻¹²). A higher intake of whole grains [β = 0.05 (SE = 0.01); P = 2.54 × 10⁻³) and seafood and plant protein [β = 0.08 (SE = 0.018); P = 1.05 × 10⁻³) was also associated with a higher relative proline betaine abundance. Intake of refined grains was inversely associated with methanol or proline, most significantly in cluster 4, [β = −0.08 (SE = 0.02); P = 7.07 × 10^−5] (Table 3).

Discussion

In this investigation, diet quality as measured by the HEI-2015 score was associated with 4 metabolites in participants in the MESA cohort study. The strongest associations were between higher HEI-2015 score and the amino acid proline betaine and an inverse association with the amino acid proline. Each cluster-associated metabolite was differentially associated with food groups. Greater intake of total fruits, whole grains, and seafood and plant protein was associated with a higher relative abundance of proline betaine. Moreover, intake of dairy products, total protein, and refined grains was associated negatively with abundance of proline.

Diet quality in the United States is low, with an mean HEI-2015 score of 59 of 100 as surveyed by NHANES in 2015–2016 [19]. Dietary intake representing high-diet quality can vary, representing broad food group categories rather than narrow associations with individual foods. Examinations of previous HEI versions have found associations between a higher HEI score and a lower risk of CVD and mortality [20,21]. This finding supports copious observational evidence that diets of high quality, generally represented by high intake of fruit, vegetable, whole grain, and plant-based protein and low intake of added sugars, salt, refined carbohydrates, and red meat, are associated with a lower incidence of chronic cardiometabolic disease [[22], [23], [24], [25]]. However, the metabolic changes and mechanisms that may underlie these associations are less clear, and the goal was to clarify representative metabolites that may indicate a high-diet quality.

A higher HEI-2015 score, representing a better diet quality, was associated with higher abundance of proline betaine. Proline betaine is also a biomarker of citrus consumption [11], reflected in this analysis with the positive association between total fruit intake and this amino acid. Similarly, in our previous work in the Mediators of Atherosclerosis in South Asians Living in America study, consumption of the fruits, vegetables, nuts, and legumes diet pattern—a high-quality diet pattern—was associated with proline betaine [26]. The DGA and most guidelines on diet intake emphasize fruit and vegetable intake as markers of high-diet quality. Because intake of fruits and vegetables likely occurs concurrently with other high-quality foods, an increase in concentration of this metabolite may serve as a general indicator for improved consumption of a high-quality diet in the general population.

Previous epidemiologic studies have shown poor cardiometabolic risk [27] and insulin resistance [28] associated with lower concentrations of betaine in diverse populations. Proline betaine and its analog, glycine betaine, were also associated with a lower risk for T2D in the Diabetes Prevention Program and other intervention and cohort studies [29,30]. In addition, deficiency of betaine was linked with increased severity of nonalcoholic fatty liver disease[31].

Betaine is derived from the amino acid glycine and acts as a methyl donor to allow the conversion of homocysteine to methionine [32]. Betaine is also a precursor of TMAO, a possible marker of cardiometabolic risk [28,33], and is likely processed by fecal microbiota into this compound. In this analysis, whole grain intake was also associated with proline betaine concentrations. In an investigation in mice, consumption of rye bran increased the diversity of gut microbiota and provided a source of glycine betaine, which was metabolized into other betaine compounds, which remained at high concentrations in the rye bran–fed group [34]. The presence of diverse microbiota from an overall healthful diet may promote higher concentrations of betaine and its metabolites throughout the gut and plasma. Despite these positive observational findings and promising preclinical data from animal studies, direct supplementation of betaine in humans during a randomized controlled trial showed only minor improvements in fasting glucose and no changes in dynamic measurements of insulin sensitivity and intrahepatic TGs [35]. Altogether, this suggests that diet intake including whole grains and cereal fiber may support a healthful gut microbial environment allowing for increasing concentrations of betaine and its metabolites, associated with lower risk for cardiometabolic disease. A deeper exploration of the choline-betaine metabolic pathways after whole grain intake may yield insights into the pathogenesis of diabetes and nonalcoholic fatty liver disease.

Total HEI-2015 score was inversely associated with the amino acid proline. Increased concentrations of proline have previously been associated with insulin resistance in South Asian and Chinese men of low BMI, suggesting that this metabolite may reflect metabolic differences underlying T2D independent of those caused by obesity [36]. This metabolite has also been inversely associated with HEI-2015 in a study of African-American and European populations [8] in an analysis restricted to known metabolites. Proline has recently been implicated in the gut–brain axis and as an indicator for the severity of depression. In a multicohort analysis, circulating proline concentration had the strongest association of all metabolites with worsened depression scores [37]. Those with high proline consumption and high plasma proline concentrations exhibited a preponderance of the gut microbiota species, such as Parabacteroides and Acidaminococcus spp. However, these gut microbiota species were also associated with higher depression scores. Because we found that a lower-diet quality was associated with higher circulating prolineconcentration, the promotion of a healthful gut environment through improved diet quality may help explain links between HEI-2015 score and depression [38].

1,5 Anhydrosorbitol (1,5 anhydroglucitol) is a marker of short-term glycemic control, inversely related to glucose concentration, and used as a validated marker of daily glucose changes. In this study, a higher HEI-2015 score was associated with lower 1,5 anhydrosorbitol concentrations. A higher intake of total dairy was similarly associated with lower circulating concentrations of this metabolite, replicating a finding in normoglycemic individuals in Japan [39]. It is readily absorbed from a variety of foods and is generally present in stable concentrations in the body because it is excreted almost without metabolism. Moreover, this metabolite was indicative of high saturated fat intake in a controlled diet trial of high saturated fat compared with n-6 FAs [40]—higher diet quality in our study is defined by lower saturated fat intake likely leading to this finding. However, circulating concentrations of this metabolite have been shown to decrease with a lower intake of overall carbohydrates or lower GI under controlled dietary intake conditions [41]. Lower concentrations of this metabolite have also been linked with an increase in major adverse cardiovascular events [42]; however, there is a stronger relationship among people with diabetes [43]. At a population level, lower intake of saturated fat and higher intake of dairy products as components of a higher HEI-2015 score may be reflected as lower 1,5 anhydrosorbitol concentrations. However, in populations with diabetes, the effect of glycemic variability on this marker likely supersedes changes from diet intake owing to competitive inhibition with glucose excretion in the renal tubules, and it is not likely to be a good indicator of diet quality in this population.

HEI-2015 score was positively associated with unsaturated fatty acyl chains (C=CHCH2HC=C). Fatty acyls are one of the 8 categories of lipids and include many different fats. The HEI-2015 component FAs, which represents the ratio of unsaturated to saturated FA intake, was associated with higher cluster 1 (unsaturated fatty acyl) score. The intake of unsaturated FAs has been linked to improved health outcomes, such as omega-3 FAs and CVD [44]. The association of higher HEI-2015 overall score to a higher ratio of unsaturated:saturated FAs was in line with expected healthy eating guidelines.

The strengths of this analysis include a longitudinal cohort design with robust habitual dietary data collection through a comprehensive FFQ, characterization of diet in multiple racial and ethnic groups, and comprehensive evaluation of untargeted NMR spectral features beyond known metabolites. Despite multiple strengths, we acknowledge that our analysis also has limitations. These findings were not externally validated, although our sample size and methods allow for adequate internal validation. This is a cross-sectional analysis performed at 1 time point, and data collected from FFQs are subject to recall bias. The FFQ data collected information on habitual diet intake over the past 12 months but did not quantify this intake at the time point of blood sampling; biomarkers may be affected by more proximate diet intake. The MESA study FFQ was modified to include unique Chinese foods and culinary practices but was not validated in this population. Untargeted metabolomics is a broad-based analysis for identifying all possible markers as a snapshot of metabolism, and this observational analysis cannot establish causal relationships between controlled diet intake and metabolites. Nevertheless, to our knowledge, our characterization of metabolites associated with HEI-2015 remains the first to broadly examine NMR spectral features associated with this dietary quality score rather than restricting the analysis to known metabolites.

In summary, the HEI-2015 score was associated with spectral features representing proline betaine, proline, 1,5 anhydrosorbitol, and fatty acyl chains in the MESA cohort study. These metabolites may represent increased whole grain, fruit, and dairy and lower saturated fat intake as indicators of overall high-diet quality. Further investigation into controlled diet intake will help to clarify links between diet quality and onset of cardiometabolic disease and areas for preventive action.

Author contributions

The authors’ responsibilities were as follows – MDG, DH: designed the research; MDG, AW: analyzed the data; MDG: wrote the paper and had primary responsibility for the final content; AW, IK, GG, IT, VWZ, PG, DH, AMK: contributed to the interpretation of the results and revised the manuscript; and all authors: have read and approved the final manuscript. This paper has been reviewed and approved by the MESA Publications and Presentations Committee.

Data availability

The data described in the manuscript, code book, and analytic code will be made available on request pending approval of the MESA cohort study.

Funding

COMBI-BIO Research was supported by EU COMBI-BIO project (FP7, 305422), and NIH/NHLBI (R01HL133932). This research was supported by 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 UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS). MDG is supported by K23DK119404 from the National Institute of Diabetes and Digestive and Kidney Diseases.

Author disclosures

ACW has received funding from Hass Avocado Board, The National Cattleman’s Beef Association, and Ionis Pharmaceuticals for work unrelated to this analyses. The other authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: