Habitual dietary intake impacts on the lipidomic profile.

Reliable dietary assessments are essential when attempting to understand the complex links between diet and health. Traditional methods for collecting dietary exposure can be unreliable, therefore there is an increasing interest in identifying biomarkers to provide a more accurate measurement. Metabolomics is a technology that offers great promise in this area. The aim of this study was to use a multivariate statistical strategy to link lipidomic patterns with dietary data in an attempt to identify dietary biomarkers. We assessed the relationship between lipidomic profiles and dietary data in volunteers (n=34) from the Metabolic Challenge Study (MECHE). Principal component analysis (PCA), linear regression and receiver operating characteristic (ROC) analysis were used to (1) reduce the lipidomic data into lipid patterns (LPs), (2) investigate relationships between these patterns and dietary data and (3) identify biomarkers of dietary intake. Our study identified a total of 6 novel LPs. LP1 was highly predictive of dietary fat intake (area under the curve AUC=0.82). A random forest (RF) classification model used to discriminate between low and high consumers resulted with an error rate of >10%, with a panel of six metabolites identified as the most predictive. LP4 was highly predictive of alcohol intake (AUC=0.81) with lysophosphatidylcholine alkyl C18:0 (LPCeC18:0) identified as a potential biomarker of alcohol consumption. LP6 had a reasonably good ability to predict dietary fish intake (AUC=0.76), with lysophosphatidylethanolamine acyl C18:2 (LPEaC18:2) phoshatidylethanolamine diaclyl C38:4 (PEaaC38:4) identified as potential biomarkers. The identification of these LPs and specific biomarkers will help in better classifying a persons dietary intake and in turn will improve the assessment of the relationship between diet and disease. Linking these LPs and specific biomarkers with health parameters will be an important future step.