Association of healthy dietary patterns and cardiorespiratory fitness in the community

Abstract

Aims

To evaluate the associations of dietary indices and quantitative cardiorespiratory fitness (CRF) measures in a large, community-based sample harnessing metabolomic profiling to interrogate shared biology.

Methods and results

Framingham Heart Study (FHS) participants underwent maximum effort cardiopulmonary exercise tests for CRF quantification (via peak VO₂) and completed semi-quantitative food frequency questionnaires. Dietary quality was assessed by the Alternative Healthy Eating Index (AHEI) and Mediterranean-style Diet Score (MDS), and fasting blood concentrations of 201 metabolites were quantified. In 2380 FHS participants (54 ± 9 years, 54% female, body mass index 28 ± 5 kg/m²), 1 SD higher AHEI and MDS were associated with 5.2% (1.2 mL/kg/min, 95% CI 4.3–6.0%, P < 0.0001) and 4.5% (1.0 mL/kg/min, 95% CI 3.6–5.3%, P < 0.0001) greater peak VO₂ in linear models adjusted for age, sex, total daily energy intake, cardiovascular risk factors, and physical activity. In participants with metabolite profiling (N = 1154), 24 metabolites were concordantly associated with both dietary indices and peak VO₂ in multivariable-adjusted linear models (FDR < 5%). Metabolites that were associated with lower CRF and poorer dietary quality included C6 and C7 carnitines, C16:0 ceramide, and dimethylguanidino valeric acid, and metabolites that were positively associated with higher CRF and favourable dietary quality included C38:7 phosphatidylcholine plasmalogen and C38:7 and C40:7 phosphatidylethanolamine plasmalogens.

Conclusion

Higher diet quality is associated with greater CRF cross-sectionally in a middle-aged community-dwelling sample, and metabolites highlight potential shared favourable effects on cardiometabolic health.

Graphical Abstract

Overview of associations of healthy dietary patterns with cardiorespiratory fitness and their joint associations with metabolites in the Framingham Heart Study. AHEI, Alternative Healthy Eating Index; MDS, Mediterranean-style Diet Score. The green color in the middle table indicates favorable changes in the specific cardiopulmonary exercise testing measures. The right scatter plot shows a representative graph of model-adjusted effects for metabolites' associations with AHEI and peak VO₂ with labeled metabolites having significant associations with peak VO₂, AHEI, and MDS. Significant associations were identified from models adjusted for age, sex, cardiovascular risk factors, and physical activity. Figure created with BioRender.com.

Lay Summary

• This study seeks to address whether healthy dietary patterns relate to gold-standard measures of physical fitness in community-dwelling adults and how circulating metabolites can demonstrate biological relationships between diet and fitness.

• Healthy diet is associated with greater physical fitness in middle-aged adults.

• The beneficial relationship between diet and fitness may be partly explained by favourable metabolic health.

Framingham Heart Study, Alternative Healthy Eating Index, Mediterranean diet, Cardiorespiratory fitness, Oxygen uptake, Metabolomics

See the editorial comment for this article ‘Association of healthy dietary patterns and cardiorespiratory fitness in the community’, by T.B. Basnet, https://doi.org/10.1093/eurjpc/zwad157.

Introduction

Cardiorespiratory fitness (CRF) is an integrative measure of physiological reserve capacity and metabolic health that is closely linked with cardiovascular outcomes.¹ While an individual’s CRF may be partially determined by standard cardiovascular disease (CVD) risk factors, there remains a large amount of inter-individual variability in CRF that is not well understood.² Lifestyle factors are also likely to contribute to CRF and may present easily modifiable avenues for improving CRF.³ Healthy dietary patterns are an important component of cardiovascular health,⁴ but it remains uncertain whether they are also related to CRF. Indeed, while putative mechanisms linking a healthy diet and improved fitness have been proposed to include anti-inflammatory effects or increased availability of antioxidants, these mechanisms are incompletely elucidated.⁵

Although association of diet with CRF has been previously reported, prior studies relied primarily on treadmill exercise time as a surrogate for peak CRF, focused on specific (often younger) populations, and used varying dietary assessments that focused on specific food types rather than standardized and integrated indices of evidence-based healthy patterns of eating.^5–14 Furthermore, few studies have examined the association of macronutrient intake with fitness in population-based cohorts. One observational study of adults presenting for preventive cardiac evaluation suggested that lower fat and higher carbohydrate intake as percentages of total daily energy were associated with higher fitness.¹³ Whether such a relation holds in community-dwelling individuals consuming more contemporary diets requires further investigation.

To address these knowledge gaps, we related indices of healthy dietary patterns in community-dwelling individuals from the Framingham Heart Study (FHS) with maximum effort cardiopulmonary exercise test (CPET) measures including peak VO₂ (the ‘gold standard’ assessment of CRF) and complementary CRF measures. We then further explored the relations between healthy dietary patterns and peak VO₂ by comparing their joint associations with >200 circulating blood metabolites. Our overarching objective was to quantify the cross-sectional relation of healthy dietary patterns with CRF measures and to gain insight into potential mechanisms through metabolite profiling. We hypothesized that healthier patterns of eating would associate with greater CRF and that metabolites associated with dietary quality and CRF would highlight potential pathways linking diet and CRF.

Methods

Study sample

Descriptions of recruitment and enrolment of the FHS Generation Three, Omni Generation Two, and New Offspring cohorts are reported.^15,16 Briefly, these cohorts were recruited together, underwent the same study protocols, and attended their first and second study visits in 2002 to 2005 and 2008 to 2011, respectively. Of the 3521 study participants attending the third study visit (2016 to 2019), 3117 participants underwent maximum effort CPET, and 2494 of these individuals also completed semi-quantitative food frequency questionnaires (FFQ). We excluded individuals with missing CPET measures (N = 18), inadequate volitional effort [peak respiratory exchange ratio (RER) < 1.05; n = 105)], and missing covariate information (n = 9), resulting in a final analytical sample of 2380 individuals (see Supplementary material online, Figure S1). From this subsample, targeted metabolomics profiling by liquid chromatography-mass spectrometry was performed on peripheral blood drawn after ≥8 h of fasting in 1154 individuals (see Supplementary material online, Figure S1). Protocols were approved by the Institutional Review Boards at Boston University Medical Campus and Massachusetts General Hospital, and all participants provided written informed consent.

Covariate assessment

We included the following covariates in our analysis: sex, age, total daily energy intake, body mass index (BMI), smoking status, total cholesterol, high-density lipoprotein cholesterol (HDL-C), systolic blood pressure (SBP), self-reported use of anti-hypertensive medication, diabetes, and physical activity index (PAI). Total daily energy intake was calculated as described below from FFQ. We classified smoking status as never, former, or current (within the last year). Diabetes was assessed as fasting blood glucose ≥126 mg/dL, non-fasting blood glucose ≥200 mg/dL, or use of glucose lowering medications. PAI was calculated based on a weighted estimate of total oxygen consumption (quantified by approximate metabolic equivalent of task) from questionnaires of typical daily activities. For sensitivity analysis, we substituted PAI with objectively measured regular physical activity based on average time spent each day sedentary or performing moderate-vigorous physical activity (MVPA) as derived from accelerometers worn for up to 8 days immediately after the study visit.³

Dietary patterns

Dietary intake was assessed using the Harvard semi-quantitative FFQ, which quantifies intake during the last year of 126 dietary items ranging from never or less than once per month to ≥6 servings/day. The FFQ has demonstrated good reproducibility and validity compared to a 1-week dietary record.¹⁷ We excluded FFQs with >12 blank items or with improbable estimated daily energy intake, defined as <600 or ≥4000 kcal/day for females and <600 or ≥4200 kcal/day for males.¹⁸ For each participant, macronutrient intake was estimated by summing frequency of consumption multiplied by nutrient composition for the specific serving size of all food items on the FFQ.¹⁹ Total daily energy intake was estimated based on 4 kcal/g for carbohydrates and proteins and 9 kcal/g for fats, and percent macronutrient intake was calculated by dividing each component’s energy contribution by the total energy.

We calculated indices of adherence to two dietary patterns, the Alternative Health Eating Index (AHEI) and the Mediterranean-style Diet Score (MDS), both of which are associated with CVD outcomes.^20,21 The AHEI aggregates a composite assessment of higher intake of vegetables, fruits, whole grains, nuts/legumes, omega-3 fatty acids, and PUFA, lower intake of sugar-sweetened beverages/fruit juice, red/processed meat, trans fatty acids, and sodium, and sex-specific moderate alcohol use (see Supplementary material online, Table S1). The AHEI assigns a score of 0 to 10 for each of the 11 dietary components and, thus, has a total range of 0 to 110 with higher scores indicating greater adherence.²⁰ To calculate the MDS, we assigned cohort- and sex-specific quartile scores (score of 0 for the lowest quartile and 3 for the highest quartile) to each of eight components: vegetables, fruits, whole grains, nuts, legumes, red meat (scored in reverse), fish, and the ratio of MUFA to SFA, and a single score for sex-specific moderate alcohol use.²² The MDS ranged from 0 to 25 with higher scores indicating greater similarity to a Mediterranean-style diet (see Supplementary material online, Table S1).

Cardiopulmonary exercise testing

Details of the CPET protocol and data collection have been described.^3,23 All participants exercised on the same cycle ergometer (Lode, Groningen, Netherlands) with breath-by-breath gas exchange values measured by the same metabolic cart (MedGraphics, St. Paul, MN). The assessment included at least 3 min of resting gas exchange measures, followed by 3 min of unloaded exercise, and then incremental ramp exercise on one of two (15 or 25 W/min) ramp protocols.³ Peak VO₂ was assessed as the highest 30-s median during the final minute of exercise. The following gas exchange and exercise variables were included: resting absolute VO₂, resting relative VO₂ (absolute VO₂ divided by body weight), peak absolute VO₂, peak relative VO₂, percent predicted peak absolute VO₂ by the Wasserman equation, resting RER, peak RER, resting heart rate (HR), peak HR, percent predicted peak HR by the Tanaka formula, VO₂/work relationship, and ventilatory efficiency (minute ventilation per 1 L/min of carbon dioxide expiration, assessed as V_E/VCO₂ nadir).²⁴

Metabolite profiling

Fasting plasma samples were collected, immediately centrifuged, and stored at −80°C without freeze-thaw cycles before assaying. Targeted measurement of polar metabolites was performed on a Nexera × 2 UHPLC (Shimadzu Corporation, Kyoto, Japan) liquid chromatograph coupled to a Q Exactive hybrid quadrupole Orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, MA) with samples injected on a 150 × 2 mm, 3-μm Atlantis hydrophilic interaction liquid chromatography (HILIC) column (Waters Corporation, Milford, MA).²³ Peak integration was performed using TraceFinder (Thermo Fisher Scientific) and identified a broad selection of metabolites based on reference samples. These metabolite classes included amino acids, amino acid metabolites, acylcarnitines, purines, pyrimidines, glycerophospholipids, and sphingolipids.

Statistical analysis

Peak VO₂ (absolute and relative) were natural log transformed for improved normality and CPET measures and dietary indices were standardized to mean = 0, SD = 1 to facilitate comparison and interpretation. Metabolite levels were inverse rank-normalized for analysis, and missing (below detection limit) metabolite values were imputed to half of the lowest detected values.²³

We evaluated the cross-sectional associations of dietary indices (AHEI and MDS; independent variables) with CPET fitness measures (dependent variables) using multivariable linear regression. First, we adjusted for age, sex, and total daily energy intake (model 1). In model 2, we additionally adjusted for BMI, smoking status, total cholesterol, HDL-C, SBP, hypertension medication use, diabetes, and PAI. Because of evidence that suggests moderate alcohol intake may not be beneficial,²⁵ we performed a sensitivity analysis excluding alcohol from AHEI and MDS. We tested for effect modification by age [<54 years (median age) vs. ≥ 54 years], sex, and BMI (<25, 25–30, or ≥30 kg/m²) on the associations between dietary patterns and key CPET variables (peak relative VO₂, resting relative VO₂, and percent predicted peak VO₂) using multiplicative interaction terms and tested the statistical significance of the interaction terms’ regression coefficients. In secondary analyses, we used the same multivariable linear regression models above to relate deciles of the percent daily energy intake from protein, fat, and carbohydrate with CPET fitness measures. A Benjamini-Hochberg false discovery rate (FDR) of 5% was used to determine statistical significance within each CPET variable and diet index analysis.

We conducted sensitivity analyses for the main associations of AHEI and MDS with CPET variables by substituting objective measures of sedentary time and MVPA for PAI in the analytic subset with available accelerometry data. We also examined the interrelationship between resting HR and peak VO₂ by adjusting each model for the other measure. We evaluated non-linearity and threshold effects for the relations of the dietary indices with log-transformed relative peak VO₂ by constructing generalized additive models that applied a smooth function to allow the AHEI or MDS to have non-linear relations with log-transformed peak VO₂; models were adjusted for age, sex, total daily energy intake, BMI, smoking status, total cholesterol, HDL-C, SBP, hypertension medication use, diabetes, and PAI. Model fits of the linear and general additive models were compared using ANOVA.

Transformed metabolite levels were then separately related to each of AHEI, MDS, and relative peak VO₂ using multivariable linear regression. Models were first adjusted for age, sex, total daily energy intake, and BMI, and then we additionally adjusted for smoking status, total cholesterol, HDL-C, SBP, hypertension medication use, diabetes, and PAI. An FDR of 5% was used to determine statistical significance separately for associations of metabolites with AHEI, MDS, and peak VO₂. We then identified metabolites significantly associated with AHEI or MDS total scores and relative peak VO₂. We related these significant metabolites to individual components of AHEI and MDS using multivariable regression, adjusted for age, sex, total daily energy intake, BMI, smoking status, cholesterol, HDL-C, SBP, hypertension medication use, diabetes, and PAI. Again, an FDR of 5% was used to determine statistical significance separately for associations of metabolites with each dietary pattern component. Analyses were performed in R 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Sample characteristics

Characteristics of the main study sample and the subsample with available metabolite measures are shown in Table 1. Participants had a mean age of 54 ± 9 years, BMI of 28.0 ± 5.3 kg/m², and peak VO₂ of 23.3 ± 6.9 mL/kg/min (91 ± 20% predicted), and 54% were female. The mean total daily energy intake was 1849 kcal/day for females and 2039 kcal/day for males. Across the cohort, overall energy intake comprised of approximately 44% carbohydrates, 36% fats, and 17% proteins. Mean AHEI and MDS were 66.7 ± 13.0 and 12.4 ± 4.7, respectively, which are largely consistent with ranges reported in other community-based studies.^20,26–28 Spearman correlation between AHEI and MDS was 0.72. Characteristics of the analytic subsample with metabolite profiling performed were similar to the main sample with a higher proportion of White participants.

Table 1

Sample characteristics

Variable	Main sample (N = 2380)	Subsample with metabolite profiling performed (N = 1154)
Age, years	54 ± 9	54 ± 8
Women	1286 (54%)	624 (54%)
Non-White race	198 (8%)	12 (1%)
Body mass index, kg/m2	28.0 ± 5.3	27.6 ± 5.2
Resting systolic blood Pressure, mmHg	119 ± 14	119 ± 14
Total cholesterol, mg/dL	190.1 ± 35.9	190.4 ± 33.6
High-density lipoprotein cholesterol, mg/dL	60.6 ± 19.3	61.6 ± 19.3
Physical activity index, METs/day	34 ± 5	34 ± 6
Smoking status
Never	1504 (63%)	735 (64%)
Former	742 (31%)	372 (32%)
Current	134 (6%)	47 (4%)
Hypertension medication use	505 (21%)	225 (20%)
Beta blocker use	178 (7.5%)	83 (7.2%)
Calcium channel blocker use	126 (5.3%)	53 (4.6%)
Diabetes	180 (7.6%)	75 (6.5%)
Prevalent cardiovascular disease	89 (3.7%)	43 (3.7%)
Alternative healthy eating index (0 to 110)	66.7 ± 13.0	66.8 ± 12.9
Mediterranean-style Diet Score (0 to 25)	12.4 ± 4.7	12.4 ± 4.7
Total daily energy intake, kcal	1936 ± 644	1959 ± 627
Energy from carbohydrates, %	44 ± 8	44 ± 8
Energy from proteins, %	17 ± 3	17 ± 3
Energy from fats, %	36 ± 7	36 ± 7
Resting absolute VO2, mL/min	280.6 ± 68.9	278.6 ± 67.2
Resting relative VO2, mL/kg/min	3.5 ± 0.6	3.5 ± 0.6
Peak absolute VO2, mL/min	1861.8 ± 638.2	1894.7 ± 656.4
Peak relative VO2, mL/kg/min	23.3 ± 6.9	23.9 ± 7.2
Percent predicted peak VO2, %	91.3 ± 20.4	92.5 ± 20.5
Resting respiratory exchange ratio	0.89 ± 0.08	0.88 ± 0.07
Peak respiratory exchange ratio	1.22 ± 0.09	1.22 ± 0.09
Resting HR, beats/min	72 ± 12	72 ± 12
Peak HR, beats/min	152 ± 19	153 ± 19
Percent predicted peak HR	89.2 ± 10.1	89.8 ± 10.1
VO2/work, mL/W/min	8.9 ± 1.0	9.1 ± 0.9
VE/VCO2 nadir	27.0 ± 2.8	26.9 ± 2.8

Numbers expressed as mean ± SD or n (%). Peak VO₂ is predicted based on the Wasserman equation. Peak heart rate was predicted using the Tanaka formula.

MET, metabolic equivalent of task.

Relations of dietary quality assessment with fitness measures

We related two different dietary quality indices (AHEI and MDS) with exercise response measures reflecting distinct physiologies. In models adjusted for age, sex, and total daily energy intake, higher dietary quality (both higher AHEI and MDS) was associated with higher resting VO₂ and with global CRF as measured by peak VO₂ (absolute, relative, and percent predicted). Dietary quality was also related to multi-dimensional exercise measures reflecting favourable autonomic function (higher peak HR and lower resting HR), greater oxygen uptake kinetics during exercise (VO₂/work), and favourable central cardiac and pulmonary vascular function (lower V_E/VCO₂; see Supplementary material online, Table S2). With further adjustment for clinical risk factors and potential confounders, we observed attenuation of the relations with resting VO₂, but associations with exercise measures remained robust (Table 2) with broad consistency across the two dietary indices. Findings were essentially unchanged when alcohol intake was excluded from the dietary assessments (see Supplementary material online, Table S3). In the multivariable adjusted models, a 1 SD higher AHEI and MDS were respectively associated with 5.2% (95% CI 4.3–6.0%) and 4.5% (95% CI 3.6–5.3%) greater peak relative VO₂ (reverse transformed based on e^{coefficient×SD log VO2} with SD of 0.296 for log peak VO₂), which translates to 1.2 and 1.0 mL/kg/min at the sample mean of 23.3 mL/kg/min. We verified the linearity of these associations with general additive models, and no threshold effects were observed (ANOVA P > 0.05 comparing linear to general additive models; Figure 1).

Figure 1

Scatterplots of peak oxygen uptake vs. Alternative Healthy Eating Index in (A) and Mediterranean-style Diet Score in (B) stratified by sex and overlaid with fitted curves from general additive models and their 95% confidence bounds. Models were adjusted for age, sex, total daily energy intake, body mass index, smoking status, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, hypertension medication use, diabetes, and physical activity index.

Table 2

Cross sectional associations of dietary patterns and cardiopulmonary exercise testing variables

CPET variables	Alternative healthy eating index (N = 2380)			Mediterranean-style Diet Score (N = 2380)
	Coefficient	SE	FDR P	Coefficient	SE	FDR P
Resting absolute VO2	0.005	0.014	0.78	0.016	0.014	0.31
Resting relative VO2	0.005	0.019	0.80	0.020	0.019	0.31
Peak absolute VO2	0.147	0.013	<0.0001	0.127	0.014	<0.0001
Peak relative VO2	0.170	0.014	<0.0001	0.148	0.015	<0.0001
Percent predicted peak VO2	0.212	0.020	<0.0001	0.183	0.021	<0.0001
Resting RER	−0.029	0.022	0.23	−0.017	0.023	0.44
Peak RER	−0.032	0.021	0.17	−0.036	0.022	0.13
Resting HR	−0.131	0.021	<0.0001	−0.115	0.022	<0.0001
Peak HR	0.055	0.018	0.0046	0.069	0.019	0.0004
Percent Predicted Peak HR	0.060	0.020	0.0049	0.077	0.021	0.0004
VO2/work	0.142	0.020	<0.0001	0.113	0.020	<0.0001
VE/VCO2 nadir	−0.089	0.020	<0.0001	−0.064	0.020	0.0026

Absolute and relative VO₂ at rest and peak were log transformed. Coefficients represent change in CPET variables (in units of SD) per one SD increase in dietary scores. Models were adjusted for age, sex, total daily energy intake, BMI, smoking status, cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, hypertension medication use, diabetes, and physical activity index. P-values were adjusted for false discovery rate using the Benjamini-Hochberg method separately for each dietary pattern.

CPET, cardiopulmonary exercise testing; FDR, false discovery rate; HR, heart rate; RER, respiratory exchange ratio.

In sensitivity analyses, we substituted objectively measured physical activity and sedentary time (available in a subset of N = 1777 individuals) for the questionnaire based PAI as adjustment variables, with minimal changes observed (see Supplementary material online, Table S4). To further explore whether the relation of a lower resting HR may be partially explained by higher fitness levels, we included both measures in models (see Supplementary material online, Table S5). Adjustment for peak relative VO₂ attenuated the effect size of dietary quality’s relations with resting HR, but they remained significant in the main model. Adjustment for resting HR had minimal influence on the effect size of the relations between dietary quality and peak VO₂, and findings were essentially unchanged after excluding 89 individuals with prevalent CVD (see Supplementary material online, Table S6).

Next, we tested for effect modification by age, sex, or BMI on the relation of dietary quality and peak VO₂ using multiplicative interaction terms. We observed a statistically significant interaction of age with both scores for the association with percent predicted peak VO₂, such that the effect estimate was higher in individuals younger than the median age of 54 years (see Supplementary material online, Table S7). There was no evidence of effect modification by sex or BMI.

Finally, we evaluated whether the intake of specific macronutrient components may be related to fitness measures in community-dwelling individuals. Consistent with known physiology and prior studies,²⁹ we observed higher carbohydrate intake to be associated with a higher resting RER (see Supplementary material online, Table S8). Additionally, we observed that after controlling for total daily energy intake, a one decile higher percent energy intake from carbohydrates was associated with lower peak VO₂, whereas a higher percent energy intake from fat was associated with higher peak VO₂.

Associations of diet and fitness measures with the circulating metabolome

To evaluate the joint relations of dietary patterns and higher CRF with blood metabolites, we evaluated linear models adjusted for clinical risk factors. In multivariable adjusted models, 83 metabolites were associated with AHEI, 81 with MDS, and 53 with peak relative VO₂ (FDR <5% for all; see Supplementary material online, Table S9). Of significant metabolite associations, 67 were shared between the two dietary patterns, 32 were shared between AHEI and peak relative VO₂, 24 were shared between MDS and peak relative VO₂, and 24 were common to all three (see Supplementary material online, Figure S2). Directions of associations of metabolites with favourable dietary quality and with higher peak VO₂ were consistent (Figure 2).

Figure 2

The beta coefficients were plotted for each metabolite in relation to peak oxygen uptake (y-axis) and dietary indices scores (x-axis) in linear models adjusted for age, sex, total daily energy intake, body mass index, smoking status, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, hypertension medication use, diabetes, and physical activity index. (A) Shows results for Alternative Healthy Eating Index, and (B) shows results for Mediterranean-style Diet Score. Point sizes reflect the average –log10 P-values for associations between peak oxygen uptake and the respective dietary index. Labelled metabolites have significant associations with respective dietary index and peak oxygen uptake after false discovery rate adjustment. For acyl group nomenclature, the number after C denotes the number of carbon atoms, and the number after the colon denotes the number of double bonds. DMGV, dimethylguanidino valeric acid; GABA, gamma-aminobutyric acid; NMMA, N-monomethyl-arginine; LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin.

Table 3 summarizes the 24 metabolites associated with both dietary patterns and peak relative VO₂. These include metabolites previously implicated in cardiometabolic health and fitness as well as novel metabolites with less well-elucidated roles in fitness. Many metabolites have established associations with cardiometabolic disease such as C6 and C7 carnitines, C16:0 ceramide, and dimethylguanidino valeric acid (DMGV), whose higher levels are linked to higher risks of diabetes and/or CVD^34,38,44 and associated with lower CRF and poorer dietary quality in our cohort. Conversely, higher C38:7 phosphatidylcholine (PC) plasmalogen and C38:7 and C40:7 phosphatidylethanolamine (PE) plasmalogen levels were associated with higher CRF and better dietary quality; as a group, plasmalogens are hypothesized to have antioxidant properties and negatively associated with metabolic diseases.³²

Table 3

Summary of metabolites associated with dietary patterns and peak relative VO₂

Metabolite class	Metabolites	AHEI/MDS	VO2	Diet and biological relationships
Glycerolipids	C34:1 DAG or TAG fragment	↓	↓	Intermediates of fatty acid metabolism, DAGs are found in small quantities (< 10%) in various seed oils65, and ingestion has been associated with decreased weight, waist circumference, and serum triglyceride65
Glycerophospho-cholines	C20:1 LPC	↑	↑	LPCs produced by cleavage of PCs by phospholipase A2, and saturated and shorter LPCs are pro-inflammatory whereas unsaturated and longer LPCs are anti-inflammatory.66,67 C22:6 LPC associated with lower risk of type 2 diabetes. C22:6 LPC positively associated with AHEI and MDS,26 fish intake,30 and vegetable intake,68 and it increases with fish oil supplementation.31 C24:0 LPC is inversely associated with diet high in meat and fast food.68
	C22:6 LPC	↑	↑
	C24:0 LPC	↑	↑
Glycerophospho-cholines	C38:7 PC Plasmalogen	↑	↑	Phospholipids found in neuronal and cardiac cell membranes. Exhibit antioxidant properties and inversely associated with obesity, diabetes, metabolic syndrome;32 C38:7 PC and C38:7 PE plasmalogens are positively associated with MDS and fish intake.30,33 Ether C38:7 PE and ether C40:7 PE increases with fish oil supplementation.31
Glycerophospho-ethanolamines	C38:7 PE Plasmalogen C40:7 PE Plasmalogen	↑	↑
Ceramides	C16:0 Ceramide d18:1	↓	↓	Associated with incident CVD,34 and inversely associated with DASH diet;35 Elevated level leads to insulin resistance, and lower levels protect from obesity in mouse models.36
Sphingomyelins	C18:1 SM	↓	↓	Higher levels associated with diabetes;37 inversely associated with AHEI, MDS, DASH diet, and fish intake26,27,35
Fatty esters	C5:1 Carnitine	↑	↑	Short chain acylcarnitine associated positively with insulin sensitivity.69
	C6 Carnitine C7 carnitine	↓	↓	Medium chain acylcarnitines associated with obesity, type 2 diabetes,38 and CVD;39 inversely associated with MDS33
Benzenes	Trimethyl-benzene	↑	↑	Used as industrial solvents and has three isomers, some of which are found naturally in plants70
Pyrimidines	Ectoine	↑	↑	Produced by various bacteria species, binds water molecules, and an osmoprotectant;71 positively associated with chicken intake30
Amino acids, peptides, and derivatives	Asparagine	↑	↑	Non-essential amino acid, associated with decreased cardiovascular and all-cause mortality72,73
	Cinnamoyl-glycine	↑	↑	Conjugate of cinnamic acid and glycine and associated with decreased risk of diabetes74; cinnamic acid is synthesized in plants, commonly found in cinnamon, and derivatives of cinnamic acid linked with improved glucose homeostasis and insulin resistance.40
	N-monomethyl-arginine (targinine)	↑	↑	Methylated form of arginine, inhibitor of nitric oxide synthase, and potent vasoconstrictor;41 associated with decreased risk of incident coronary heart disease in African Americans.42
	N-acetyl-ornithine	↑	↑	Intermediate metabolite in the biosynthesis of arginine; positively associated with tea intake, DASH diet, vegetable intake, and AHEI30,35,43
	N-acetyl-tryptophan	↑	↑	Inhibitor of cytochrome c release and the binding of substance P to neurokinin 1 receptor with potential downstream neuroprotective effects75
	Pantothenate (vitamin B5)	↑	↑	Component of coenzyme A synthesis, essential for fatty acid metabolism; positively associated with MDS26
	Pipecolic acid	↑	↑	Intermediate metabolite in the metabolism of lysine to 2-aminoadipic acid, which is associated with increased risk of incident diabetes;76 elevated in peroxisome diseases, reported to have an inhibitory effect on the central nervous system;77 positively associated with MDS and DASH diet33,35
	Valine	↑	↑	A branched chain amino acid, related to increased risk of incident diabetes and insulin resistance;78,79 metabolized to beta-amino-isobutyric acid, which induces brown adipocyte specific gene expression80
Carboximidic acids	N-carbamoyl-beta-alanine (ureidopropionic acid)	↑	↑	Intermediate metabolite of uracil catabolism to beta-alanine
Keto acids	Dimethyl-guanidino valeric acid (DMGV)	↓	↓	Associated with incident diabetes, non-alcoholic fatty liver disease, coronary artery disease, and cardiovascular mortalit44,45; positively associated with sugar-sweetened beverage intake and inversely associated with vegetable intake44

Number after C denotes carbon atoms in acyl side chains and number after colon denotes number of double bonds. References^65–80 are included in the Supplementary material online.

AHEI, alternative healthy eating index; DAG, diacyclglycerol; DASH, dietary approaches to stop hypertension; LPC, lysophosphatidylcholine; MDS, Mediterranean-style Diet Score; PC, phosphatidylcholine; PE, phosphatidylethanolamine; TAG, triacylglycerol.

We then further explored the relations of metabolites with individual components of each dietary index (Figure 3). Some of the strongest metabolite associations with individual food groups included glycerophospholipids such as C40:7 and C38:7 PE plasmalogens, C38:7 PC plasmalogen, C22:6 lysophosphatidylcholine (LPC), C20:5 LPC, and C38:4 PE, which were positively associated with fish and omega-3 fatty acid intake, consistent with prior work.^30,31 These findings further support the validity of our dietary assessments by demonstrating relations with biochemical signatures of their intake.

Figure 3

Heat maps of the estimated regression coefficients from models relating metabolites that were significantly related to peak oxygen uptake and Alternative Healthy Eating Index and Mediterranean-style Diet Score to dietary index components are displayed. Linear regression models adjusted for age, sex, total daily energy intake, body mass index, smoking status, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, hypertension medication use, diabetes, and physical activity index. For acyl group nomenclature, the number after C denotes the number of carbon atoms, and the number after the colon denotes the number of double bonds. HILIC, hydrophilic interaction liquid chromatography; NMMA, N-monomethyl-arginine; GABA, gamma-aminobutyric acid; DMGV, dimethylguanidino valeric acid; PE, phosphatidylethanolamine; PC, phosphatidylcholine; LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; SM, sphingomyelin; EPA, eicosapentaenoic acid; DHA, docosahexaenoic acid; MUFA, monounsaturated fatty acid; SFA, saturated fatty acid.

Discussion

Here, we provide a comprehensive examination of the associations of healthy dietary patterns with fitness measures in a large, community-based sample and quantify a broad circulating metabolome to explore joint associations of diet and fitness. Two readily applicable healthy dietary patterns were associated with CRF in our study with broad consistency across age, sex, and BMI categories. Specifically, a 1 SD higher AHEI or MDS was associated with a ≈5% greater peak VO₂ (1.2 mL/kg/min at the sample mean), a magnitude similar to that observed for taking ≈4000 more steps per day in our previous work.³ Adherence to healthy dietary patterns was also associated with multi-dimensional fitness measures reflecting distinct physiologies including autonomic function (rest and exercise HR), central cardiac and pulmonary vascular function (V_E/VCO₂),²⁴ and oxygen uptake kinetics (VO₂/work). By merging metabolite associations with dietary indices and CRF, we observed broad directional consistency between metabolite relations with favourable dietary quality and CRF, highlighted metabolites previously implicated in cardiometabolic disease, and identified novel metabolites that may link the shared biology of diet and fitness.

Prior studies have demonstrated associations of dietary components with fitness but have primarily focused on specific populations (e.g. extremes of age),^6–10 estimated measures of fitness,^5,6,11–14 or specific dietary components.^6,10,12,13 Our study adds to the literature by directly addressing some of the limitations of prior investigations on dietary patterns and fitness. First, the use of CPET enabled quantification of peak VO₂ for objective assessment of fitness and ensured that an adequate (i.e. diagnostic) level of volitional effort was provided by all participants (via the RER). By using comprehensive CPET, we were also able to measure the association of diet with different physiological components of fitness. For example, we observed the expected relation of higher carbohydrate intake with resting RER,²⁹ which served as a positive control and supported the accuracy of our dietary indices. The relations of dietary quality with measures of autonomic (HR responses) and central cardiac/pulmonary vascular function (VE/VCO₂) with exercise may also indicate that dietary quality has broader salutary effects beyond fuel substrate utilization on various organ systems.

Second, our large community-based sample enabled assessment of the relations between dietary quality and fitness throughout a broad age range and how they varied across sex, age, and BMI. We observed consistent associations (i.e. lack of effect modification) across sex and BMI categories, but there was a statistically significant interaction between age and dietary quality scores on the association with percent predicted peak VO₂, suggesting a higher effect size in younger individuals. This finding is consistent with prior studies that have often demonstrated associations between dietary quality and fitness in younger cohorts.^6,9,10

Third, we captured dietary patterns with a widely used FFQ and two healthy dietary indices designed to reflect the best contemporary evidence for healthy eating.^20,21 Both the AHEI and MDS have been robustly associated with lower risk of cardiovascular and all-cause mortality and other health outcomes,^20,21,46,47 and their components/scores are readily translatable to individuals or for use in intervention studies. Furthermore, the use of dietary patterns captures synergies between food sources and better reflects how individuals typically eat. We observed that as percentages of total daily energy consumed, increased carbohydrate intake, and decreased fat intake were associated with lower peak VO₂. At the extremes, a very low-carbohydrate or ketogenic diet has been popularized for enhancing aerobic performance based on the theory that ketones may be more optimal fuel sources; however, the data for ketogenic diet improving peak VO₂ in athletes are mixed.⁴⁸ A low-fat (<30% total daily energy intake) diet has been historically recommended for population health,⁴⁹ but studies have increasingly shown that the quality and source of dietary fats and carbohydrates matter, leading to a focus on dietary patterns in recent guidelines.⁵⁰ In contrast to our findings, another observational study showed that lower fat intake was associated with higher CRF, but men in the study with high fitness also consumed less SFA (10.0% vs. 11.8% of total daily energy intake) and MUFA (12.6% vs. 14.5%) and a similar amount of PUFA (7.4% vs. 7.4%) compared to those with low fitness.¹³ This discordance highlights the importance of food sources when considering macronutrient intake.

These epidemiological observations highlight the need for further investigation of potential mechanisms linking diet and CRF. To begin to address this gap, we explored potential shared biological processes by leveraging metabolite profiling. We identified 24 metabolites that were significantly associated with peak VO₂, AHEI, and MDS after adjustment for physical activity and standard CVD risk factors. A particular strength of metabolomic interrogation in dietary research is that some metabolites may provide objective measures of the consumption of specific food items or the host-diet interaction. Indeed, several of our metabolite-dietary pattern associations have been identified by prior studies.^{26,27,33,35,43,51,52} Among our significant metabolites, those with previously reported positive associations with AHEI and MDS included lipid species such as C22:6 LPC, C38:7 PC plasmalogen, and C38:7 and C40:7 PE plasmalogens, which were all positively associated with greater fish intake. This finding is biochemically consistent with DHA being the fatty acid side chain in C22:6 LPC. Although our metabolomics platform did not precisely determine the fatty acid side chain composition of C38:7 PC plasmalogen, and C38:7 and C40:7 PE plasmalogens, DHA is likely one of their side chains based on their most plausible molecular structures. Omega-3 fatty acids are known to have favourable health benefits, and plasmalogens are hypothesized to have antioxidant properties and are inversely correlated with metabolic diseases.³² It remains unclear whether these metabolites have benefits on CRF themselves vs. being markers of omega-3 fatty acid intake. In small placebo-controlled pilot studies of athletes and sedentary adults, supplementation with DHA and EHA improved submaximal exercise HR, submaximal VO₂, and/or peak VO₂,^53–55 and in rat models, dietary omega-3 fatty acids reduced myocardial oxygen consumption while maintaining external work.⁵⁶ Potential mechanisms include increased altered mitochondrial membrane composition and mitochondrial biogenesis.^57,58

Metabolites with known inverse associations with AHEI and MDS included C6 and C7 carnitines, C18:1 sphingomyelin (SM), and C16:0 ceramide, which were also positively associated with intake of red meat, trans-fatty acid, and SFA. Medium chain carnitines such as C6 and C7 carnitines are positively associated with obesity, type 2 diabetes, and CVD,^38,39 and shorter chain sphingomyelins are positively associated with incident diabetes.³⁷ Ceramides have been recently identified as markers of incident CVD³⁴; in mouse models, elevated C16:0 ceramide led to insulin resistance, and decreased C16:0 ceramide protected against diet-induced obesity.³⁶ Consequently, these metabolites are possibly linked to poor CRF through their relation with insulin resistance, which has been hypothesized to induce mitochondrial dysfunction⁵⁹ and may contribute to exercise intolerance.

In addition to the above known diet-related metabolites, we identified several metabolites displaying novel associations with dietary patterns and peak VO₂. One notable example is N-monomethyl-arginine (NMMA), which inhibits nitric oxide synthase and causes increased vascular resistance.⁴¹ Given this mechanism, its positive association with peak relative VO₂ was surprising and may reflect improved autoregulation of blood pressure in those with higher CRF, whereas those with chronically increased vascular resistance may have suppressed NMMA levels. In a randomized cross-over study of 10 healthy volunteers, the administration of L-NMMA increased resting blood pressure but did not change peripheral vascular resistance or cardiac output during or after exercise,⁶⁰ and in a population study of African Americans, higher NMMA was associated with decreased risk of incident coronary artery disease.⁴² Also related to the nitric oxide pathway, we observed significant positive associations of AHEI, MDS, and peak relative VO₂ with N-acetyl-ornithine, which is an intermediate metabolite for arginine synthesis.

Overall, our analysis of metabolite associations with CRF and diet also implicated several metabolites with previously reported associations with cardiometabolic diseases. For example, higher cinnamoylglycine was associated with healthy dietary patterns and better CRF and was reported to be associated with decreased diabetes risk and improved insulin resistance.⁴⁰ By contrast, higher DMGV was associated with poorer dietary patterns and lower CRF and has been linked to poor metabolic health.⁴⁵ Given that CRF reflects underlying metabolic health, further studies are needed to determine whether the above metabolites are merely markers of dysmetabolism or whether they may have functional roles during exercise.

Our study provides a comprehensive assessment of the cross-sectional association between dietary quality and CRF in a large community-dwelling sample and uses their associations with the circulating metabolome to infer potential biological mechanisms. Nevertheless, there are several limitations that warrant consideration. Despite adjustment for several potential confounders, we cannot exclude the possibility of residual confounding or reverse causality in this cross-sectional analysis; in particular, it is possible that individuals with higher CRF may choose to live healthier lifestyles with better diets. Future interventional studies are necessary to clarify whether the associations between diet and CRF are indeed causal. Moreover, how dietary choices at different time intervals prior to exercise (e.g. immediately before, in the days before, or in the weeks/months before) might influence CRF remains uncertain and would require prospective trials. Dietary quality assessment relied on self-recall and therefore may have limitations in regard to validity and reproducibility. While we do have several positive controls in our study that attest to the accuracy of the dietary information collected (e.g. peak RER and carbohydrate association, individual metabolites reflecting known food sources), the possibility for the measurement error of dietary quality cannot be excluded. Moreover, we selected two healthy dietary patterns that have been associated with favourable cardiovascular health, but our findings may not be generalizable to all healthy dietary patterns. Future studies are needed to evaluate specific dietary components that may influence CRF directly and whether supplementation with ergogenic substances (e.g. metabolites themselves) might lead to improved exercise performance.⁶¹ Our analyses focused on peak VO₂ as an assessment of CRF given its clinical applicability, but this value may underestimate the true maximum VO₂.⁶² Additionally, while cycle exercise may lead to slightly lower peak VO₂ values when compared to other exercise modalities (e.g. treadmill),⁶³ peak VO₂ assessments obtained via different exercise modalities are highly correlated and would therefore be expected to result in similar effect estimates.⁶⁴ Finally, despite the inclusion of the multi-ethnic Omni cohort, our study sample mostly consisted of White individuals eating American-style diets; the generalizability to other racial and ethnic groups is, therefore, unknown.

In conclusion, we showed in a large observational study of community-dwelling individuals that a healthy diet is associated with greater CRF and other physiologically relevant exercise responses. We identified several metabolites that were associated with both healthy dietary patterns and peak VO₂, and many of these metabolites are known markers of cardiometabolic health. These findings provide a foundation for future work investigating how improved metabolic health may link diet and CRF.

Supplementary material

Supplementary material is available at European Journal of Preventive Cardiology.

Author contributions

M.Y.M., P.G., M.E.W., V.L.M., R.S.V., R.V.S., G.D.L., and M.N. designed research. M.E.W., P.M., V.X., R.S.V., R.V.S., G.D.L., and M.N. conducted research. M.Y.M., P.G., M.E.W., V.L.M., M.G.L., R.S.V., R.V.S., G.D.L., and M.N. analyzed data. M.Y.M., P.G., and M.N. wrote paper. M.N. had primary responsibility for final content. All authors have read and approved the final manuscript.

Funding

The Framingham Heart Study is supported by the National Heart, Lung, and Blood Institute at the National Institutes of Health (contracts N01-HC-25195, HHSN268201500001I, and 75N92019D00031). M.Y.M. is supported by the National Institutes of Health (grant number T32-HL007208). M.E.W. is supported in part by the American Heart Association (grant number 20CDA35310237), the Doris Duke Charitable Foundation (grant number 2021261), and the National Center for Advancing Translational Sciences at the National Institutes of Health through BU-CTSI (grant number 1UL1TR001430). R.S.V. was supported in part by the Evans Medical Foundation and the Jay and Louis Coffman Endowment from the Department of Medicine, Boston University School of Medicine. R.V.S. is supported by the National Institutes of Health (grant number R01-HL136685). G.D.L. is supported by National Institutes of Health (grant number R01-HL131029) and American Heart Association (grant number 15GPSGC24800006). M.N. is supported by National Institutes of Health (grant number K23-HL138260, R01-HL156975) and by a Career Investment Award from the Department of Medicine, Boston University School of Medicine.

Data availability

Data from the Framingham Heart Study are publicly available through requests via the National Institutes of Health database of Genotypes and Phenotypes (https://www.ncbi.nlm.nih.gov/gap/).

Previous presentation: Data in this paper were presented at the 2022 American Heart Association’s Annual Scientific Sessions in Chicago, IL.

Disclaimer: The contents of this work are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health.

Author notes

Conflict of interest: V.L.M. owns stock in Amgen, General Electric, and Cardinal Health. He has received speaking honoraria from, serves as a scientific advisor for, and owns stock options in Ionetix. He has received research funding and speaking honoraria from Siemens Medical Imaging. He has served as a scientific advisor for Curium and has received expert witness fees from Jubilant Draximage. He has received a speaking honorarium from 2Quart Medical. He has received non-financial research support from INVIA Medical Imaging Solutions. In the past 12 months, R.V.S. has served as a consultant for Myokardia, Best Doctors, Amgen, and Cytokinetics. R.V.S. is a co-inventor on a patent for ex-RNAs signatures of cardiac remodelling. The spouse of R.V.S. works for UpToDate (Wolters Kluwer). G.D.L. has research funding from Amgen, Cytokinetics, Applied Therapeutics, AstraZeneca, and Sonivie in relation to projects and clinical trials investigating exercise capacity that are distinct from this work. He has served as a scientific advisor for Pfizer, Merck, Boehringer-Ingelheim, Novartis, American Regent, Relypsa, Cyclerion, Cytokinetics, and Amgen and receives royalties from UpToDate for scientific content authorship related to exercise physiology. M.N. has received speaking honoraria from Cytokinetics.