Improving Precision in Estimating Diet–Disease Relationships With Metabolomics

Andrew Mente; Philip Britz-McKibbin; Salim Yusuf


Eur Heart J. 2023;44(7):570-572. 

In This Article

Abstract and Introduction


Graphical Abstract

Schematic diagram conceptualizing the pathways between diet and clinical events and the use of diverse complementary methods to provide robust answers on diet and health.


In the 1950s and 1960s, with the increase in coronary heart disease in Western countries, diet became a major focus of research in cardiovascular epidemiology.[1] This included the effects of nutrients (especially saturated fats) and certain foods (red meat, dairy, and eggs). The focus on fats was based on the idea that reducing saturated fat will reduce a single intermediate risk factor (i.e. LDL cholesterol) which in turn would presumably reduce the risk of heart disease.[2] Yet low-fat or fat-free food products were simply substituted with a higher carbohydrate content to retain palatability, highlighting the imprudence of such 'heart-healthy' dietary recommendations that focus on a single class of macronutrient. Subsequent work showed that a reliance on intermediate risk markers does not reliably predict the effects of diet on disease.[3,4] Observational cohort studies have generally shown inconsistent findings, with most studies showing no significant association between meat consumption and cardiovascular disease (CVD), and even protective associations between dairy and CVD (modest risk reduction of 5–15%).[5–7]

Some of the challenges of dietary studies include modest strength of associations and a reliance on surrogate markers (e.g. LDL cholesterol or blood pressure) which have been inconsistent at predicting net clinical impact. Because large randomized trials of diet and clinical outcomes are challenging to conduct, the field has mainly relied on cohort studies or short-term randomized studies of the effects of diets on intermediate measures such as blood pressure, or lipids. In most large observational studies, diet is self-reported, which often leads to random measurement errors in estimating diet intake and can therefore dilute associations of diet and disease. Additionally, epidemiological studies are subject to confounding even after extensive covariate adjustments, especially if the observed effects are modest, which is the case with most studies of diet. These factors mean that results from observational studies or randomized trials of intermediate markers cannot by themselves adequately inform policies on diet and health.

Nutritional metabolomics is a new approach that has been introduced in epidemiological studies, and may help in more objectively measuring food exposures while discovering new intermediate biomarkers linking diet to CVD.[8] By simultaneously triangulating and comparing dietary sources of blood biomarkers and clinical outcomes, one can determine whether specific biomarkers provide better prediction of CVD events. Also, this strategy can elucidate the specificity of circulating metabolites with habitual food intake that may be generalizable in different populations. Such studies hold promise to provide deeper knowledge of the pathways from diet to CVD. This could open up new approaches to chronic disease prevention which could then be evaluated in randomized trials.

In their study published in this issue of the European Heart Association, Shah et al.[9] examined >4000 (only 2259 analysed with metabolomics) individuals in two US-based cohort studies to derive and validate multi-metabolite signatures of dietary pattern against CVD. First, the authors examined 17 food groups, 2 nutrient groups, and healthy eating index-2015 (HEI2015) diet quality score, and identified an array of metabolites and lipids associated with diet, including food-related components (e.g. fish and long-chain unsaturated triacylglycerols), interactions with host features (microbiome), and pathways broadly implicated in cardiometabolic–cardiovascular disease (CM-CVD) (e.g. ceramide/sphingomyelin lipid metabolism). The authors then used machine learning models to define circulating metabolites linked to dietary pattern classically unfavourable for CM-CVD (e.g. high in red/processed meat, refined grains, and sugary drinks; low in vegetables, whole grains, nuts, and fruit). Metabolic signatures reflecting this pattern were associated with long-term diabetes and CVD risk more strongly than HEI2015 in CARDIA (e.g. diabetes, hazard ratio 1.62; CVD, hazard ratio 1.55 compared with 1.31 and 1.34 with diet questionnaires), with associations replicated for diabetes (P < 0.0001) but not for CVD in the Framingham Heart Study. The authors concluded that metabolomics improves precision to estimate dietary effects on adverse health outcomes.

The two health cohorts have several important strengths which include a prospective design with a prolonged follow-up, careful documentation of habitual dietary intake, and adjustment for known confounders (including other dietary components). The replication of these results using two separate US cohorts with different age, race, and ethnicity, is reassuring, but studies from populations from other regions of the world where diets differ will be needed to assess the generalizability of the observations. As with any observational study, residual confounding is a possibility, and no degree of statistical adjustments can completely account for such effects, although the authors have done as good a job as any.

While epidemiological studies have limits in exploring associations between diet and disease, the use of metabolomics methods has its own challenges.[8] The development of a comprehensive 'metabolite profile' is expensive and, once the findings are replicated, further work into the mechanisms behind the observations is needed. Most biomarkers for CM-CVD have not been confirmed as being causal, with significant variations in host metabolism and microbiome activity that impact their circulating concentrations; comprehensive biomarker screening is expensive and thus prohibitive on a large scale, and most metabolites are not reflective of long-term food intake. These limitations can now be partly overcome through advances in multiplex biomarker assays and higher throughput mass spectrometry-based platforms, which can examine hundreds of biomarkers at a relatively modest cost per participant. Recently, several observational cohort studies (Nurses' Health Study and Health Professionals Follow-Up Study) and the PREDIMED (Prevención con Dieta Mediterránea) trial derived a composite metabolite score based on the Mediterranean diet pattern that was associated with CVD. However, while the Mediterranean diet is believed to be healthy, it is not necessarily common in many other parts of the world, and Mediterranean diets may be associated with other health behaviours and environmental factors such as social cohesion and support.

The finding in this study that the metabolite-based diet intake scores have stronger associations with CM-CVD, in comparison with diet survey-based diet measures, may reflect either greater measurement error in quantifying diet by survey (which is expected to dilute associations) or the ability of the metabolome to capture disease risk beyond dietary exposures, or both. In the authors' analysis of the Framingham Heart Study, estimates of association with CM-CVD were diluted after adjustment for risk factors, which suggests that intermediates (diabetes, hypertension, and hyperlipidaemia) may mediate associations between diet–disease metabolites and CM-CVD. Larger studies in diverse populations having distinct eating patterns are needed to explore this finding further.