Little Contribution of Genome-Wide Gene-Diet Interactions to Cardiometabolic Risk Factors in the UK Biobank

Abstract

ObjectivesGenome-guided dietary recommendations have long been a goal of precision nutrition and have a substantial evidence base in well-controlled animal experiments. However, it remains unclear how much these gene-diet interactions (GxDs) affect disease risk in human populations. Understanding the contribution of GxDs to variability in disease risk factors is thus critical to guide scientific effort as well as prioritize specific dietary exposures for study. We set out to quantify the genome-wide contribution of GxDs to the heritability of multiple cardiometabolic risk factors across a wide range of dietary traits.MethodsThe study population consisted of unrelated, European-ancestry individuals from the UK Biobank with dietary data available from one or more 24-hour recall questionnaires (N = 122,945). We conducted a genome-wide GxD interaction study for each combination of 12 cardiometabolic risk factors (e.g., lipids and glycemic traits) and 20 dietary exposures (macronutrients, micronutrients, and food groups). We then used the resulting genome-wide summary statistics as input to the regression-based GxEsum method. GxEsum estimates the contribution of genetic main effects or interaction effects to heritability, expressed as a fraction of the outcome trait variability with associated confidence intervals.ResultsNo diet-trait pair passed a Bonferroni significance threshold, though 13 pairs were nominally significant (p < 0.05). The strongest of these was the gene-by-whole grain interaction, explaining 1.1% of the variability in systolic blood pressure (p = 0.002). GxD contributions to heritability (mean of 0.1% variance explained across all diet-trait pairs) were much lower than those from genetic main effects (mean of 16.8% variance explained). Finally, we saw substantial attenuation of heritability estimates when regressions adjusted for GxD terms for the remaining dietary components; this reinforces conclusions from non-dietary studies and indicates the importance of adjusting for “interaction confounders” in nutrigenetic studies.ConclusionsOur results show that it is difficult to find meaningful GxD signal for common traits, even in a large and homogeneous population. This may indicate a true lack of GxD effects or a need for improved (e.g., longitudinal or biomarker-based) dietary behavior estimation.Funding SourcesNHLBI.