Identification and validation of poly-metabolite scores for diets high in ultra-processed food: An observational study and post-hoc randomized controlled crossover-feeding trial.

Abstract

Background: Ultra-processed food (UPF) accounts for a majority of calories consumed in the United States, but the impact on human health remains unclear. We aimed to identify poly-metabolite scores in blood and urine that are predictive of UPF intake.

Methods and findings: Of the 1,082 Interactive Diet and Activity Tracking in AARP (IDATA) Study (clinicaltrials.gov ID NCT03268577) participants, aged 50-74 years, who provided biospecimen consent, n = 718 with serially collected blood and urine and one to six 24-h dietary recalls (ASA-24s), collected over 12-months, met eligibility criteria and were included in the metabolomics analysis. Ultra-high performance liquid chromatography with tandem mass spectrometry was used to measure >1,000 serum and urine metabolites. Average daily UPF intake was estimated as percentage energy according to the Nova system. Partial Spearman correlations and Least Absolute Shrinkage and Selection Operator (LASSO) regression were used to estimate UPF-metabolite correlations and build poly-metabolite scores of UPF intake, respectively. Scores were tested in a post-hoc analysis of a previously conducted randomized, controlled, crossover-feeding trial (clinicaltrials.gov ID NCT03407053) of 20 subjects who were admitted to the NIH Clinical Center and randomized to consume ad libitum diets that were 80% or 0% energy from UPF for 2 weeks immediately followed by the alternate diet for 2 weeks; eligible subjects were between 18-50 years old with a body mass index of >18.5 kg/m2 and weight-stable. IDATA participants were 51% female, and 97% completed ≥4 ASA-24s. Mean intake was 50% energy from UPF. UPF intake was correlated with 191 (of 952) serum and 293 (of 1,044) 24-h urine metabolites (FDR-corrected P-value < 0.01), including lipid (n = 56 serum, n = 22 24-h urine), amino acid (n = 33, 61), carbohydrate (n = 4, 8), xenobiotic (n = 33, 70), cofactor and vitamin (n = 9, 12), peptide (n = 7, 6), and nucleotide (n = 7, 10) metabolites. Using LASSO regression, 28 serum and 33 24-h urine metabolites were selected as predictors of UPF intake; biospecimen-specific scores were calculated as a linear combination of selected metabolites. Overlapping metabolites included (S)C(S)S-S-Methylcysteine sulfoxide (rs = -0.23, -0.19), N2,N5-diacetylornithine (rs = -0.27 for serum, -0.26 for 24-h urine), pentoic acid (rs = -0.30, -0.32), and N6-carboxymethyllysine (rs = 0.15, 0.20). Within the cross-over feeding trial, the poly-metabolite scores differed, within individual, between UPF diet phases (P-value for paired t test < 0.001). IDATA Study participants were older US adults whose diets may not be reflective of other populations.

Conclusions: Poly-metabolite scores, developed in IDATA participants with varying diets, are predictive of UPF intake and could advance epidemiological research on UPF and health. Poly-metabolite scores should be evaluated and iteratively improved in populations with a wide range of UPF intake.