Relating Urinary Polyphenol Metabolite Profiles To Specific Polyphenol-Rich Foods In The European Prospective Investigation Into Cancer And Nutrition (EPIC) Study

Hwayoung NOH, International Agency for Research on Cancer , France
FREISLING H. 1 , ASSI N. 1 , ZAMORA R. 1 , ACHAINTRE D. 1 , SLIMANI N. 1 , SCALBERT A. 1 , FERRARI P. 1

1 Nutrition and Metabolism Section, International Agency for Research on Cancer (IARC), Lyon, France

Purpose: Metabolomics offers great potential to improve the accuracy of dietary intake measurements in nutritional epidemiology. This study aimed to develop an analytical framework to use profiles of urinary polyphenols to predict intake of polyphenol-rich foods, applying a multivariate statistical technique, reduced rank regression (RRR), in the EPIC cross-sectional study.
Methods: This study included 475 subjects aged 35-70 years randomly selected from 4 European countries. Dietary data were collected using 24-hour dietary recall (24-HDR) and dietary questionnaires (DQ). Thirty-four urinary polyphenols were measured by UPLC-ESI-MS-MS in 24-hour urines, collected the same day of the 24-HDR interview. RRR analyses identified linear combination of polyphenols to maximize the explained variability of specific foods and food groups. To evaluate the performance of RRR models, cross-validation analyses were conducted by splitting the data into a training and a test set, and computing correlation coefficients and area under curve (AUC) statistics to discriminate between consumers and non-consumers.
Results: The RRR scores of polyphenol profiles were correlated with intakes of red wine (24HDR radjusted=0.67, DQ radjusted=0.22), citrus fruit (24HDR radjusted =0.62, DQ radjusted=0.21) and coffee (24HDR radjusted=0.61, DQ radjusted=0.54). Highest predicting performance was observed for coffee (ROC AUC=91.4%), red wine (87.6%) and citrus fruits (85.0%) from 24-HDR. Correlation coefficients and ROC AUC values were consistently higher for 24-HDR than for DQ. The RRR models in the test set also well predicted intakes of coffee (AUC=85.0%), red wine (79.9%), and citrus fruits (76.6%) from 24-HDR.
Conclusion: RRR lent itself as a useful tool to identify linear combinations of polyphenols that can predict intake of specific food groups, especially from 24-HDR. Compared to single food/metabolite comparisons, multivariate analyses provide better predictive performance, and a promising strategy to integrate available metabolomic and dietary information.
Funding source: EU-financed FP7 project (NutriTech grant no.289511), NA funded by EDISS, Université de Lyon.