Identifying the metabolomic fingerprint of high and low flavonoid consumers

Abstract

High flavonoid consumption can improve vascular health. Exploring flavonoid-metabolome relationships in population-based settings is challenging, as: (i) there are numerous confounders of the flavonoid-metabolome relationship; and (ii) the set of dependent metabolite variables are inter-related, highly variable and multidimensional. The Metabolite Fingerprint Score has been developed as a means of approaching such data. This study aims to compare its performance with that of more traditional methods, in identifying the metabolomic fingerprint of high and low flavonoid consumers. This study did not aim to identify biomarkers of intake, but rather to explore how systemic metabolism differs in high and low flavonoid consumers. Using liquid chromatography-tandem MS, 174 circulating plasma metabolites were profiled in 584 men and women who had complete flavonoid intake assessment. Participants were randomised to one of two datasets: (a) training dataset, to determine the models for the discrimination variables (n 399); and (b) validation dataset, to test the capacity of the variables to differentiate higher from lower total flavonoid consumers (n 185). The stepwise and full canonical variables did not discriminate in the validation dataset. The Metabolite Fingerprint Score successfully identified a unique pattern of metabolites that discriminated high from low flavonoid consumers in the validation dataset in a multivariate-adjusted setting, and provides insight into the relationship of flavonoids with systemic lipid metabolism. Given increasing use of metabolomics data in dietary association studies, and the difficulty in validating findings using untargeted metabolomics, this paper is of timely importance to the field of nutrition. However, further validation studies are required.

Keywords: Diet; Epidemiology; Flavonoids; HPFS, Health Professionals Follow-Up Study; Metabolomics; NHS, Nurses’ Health Study; Population; QC, quality control.

Fig. 1.

Study design.

Fig. 2.

Multivariate-adjusted metabolome-wide association study of flavonoid intake and the twenty metabolites to which it is most strongly associated. -----, Bonferroni-corrected level of significance required, after accounting for 174 multiple comparisons. Multivariate-adjusted model includes case/control status, cohort, quintiles of energy intake, smoking status, age at blood collection, the Alternative Healthy Eating Index (minus alcohol) score and alcohol consumption.

Fig. 3.

ANCOVA of the Metabolite Fingerprint Score by flavonoid intake group in the training dataset. Results are least squared mean values, with their standard errors represented by horizontal bars. * Significantly different from low consumers (P < 0·05). The multivariate-adjusted model includes case/control status, cohort, quintiles of energy intake, smoking status, age at blood collection, the Alternative Healthy Eating Index (minus alcohol) score and alcohol consumption. Low intake, n 123; moderate intake, n 147; high intake, n 129.

Fig. 4.

ANCOVA of the Metabolite Fingerprint Score by flavonoid intake group in the validation dataset. Results are least squared mean values, with their standard errors represented by horizontal bars. * Significantly different from low consumers (P < 0·05). The multivariate-adjusted model includes case/control status, cohort, quintiles of energy intake, smoking status, age at blood collection, the Alternative Healthy Eating Index (minus alcohol) score and alcohol consumption. Low intake, n 60; moderate intake, n 67; high intake, n 58.