Food consumption significantly impacts health and contributes to excess mortality globally. In Western societies, readily available food has shifted dietary behavior from mere consumption to choice, making understanding the drivers of food choice crucial for promoting healthier diets. While food preference is complex, encompassing personal preferences, health, ethics, and socioeconomic factors, food liking reflects the hedonic response and is more closely tied to biology. Understanding food liking is critical for designing effective dietary interventions and creating more appealing nutritious foods. Food liking is a complex trait influenced by genetics, biology, psychology, environment, branding, and culture. Twin studies show moderate heritability, with around 50% of variance in children explained by genetics and shared environmental effects. In adults, heritability remains stable but shared environmental influence diminishes. While GWAS have investigated genetic variants associated with food consumption, studies on food liking have mostly been limited to candidate gene studies with mixed results. Recent genome-wide approaches have identified some genes related to liking of specific foods or tastes, but these studies often lack power due to small sample sizes. This study uses a large sample from the UK Biobank to perform a GWAS on food liking and explore its genetic architecture, examining relationships between different food preferences, and exploring genetic correlations with other traits.

Existing research highlights the significant influence of genetic and environmental factors on food preferences and consumption patterns. Twin studies consistently demonstrate moderate heritability of food preferences, with the relative contribution of shared and non-shared environmental factors varying across age groups. Previous genetic association studies have focused primarily on candidate genes related to taste receptors, with varying degrees of success in identifying robust associations with specific food likings. While some studies have utilized genome-wide approaches, their sample sizes often limited the power to detect modest effects of common genetic variation on more specific food-liking traits. A need for a larger-scale, comprehensive GWAS examining a wide range of food items was identified to provide a more robust understanding of the genetic architecture of food liking and its relationship with other traits.

This genome-wide association study (GWAS) was conducted using data from the UK Biobank, including over 150,000 participants of European descent who completed a food liking questionnaire. The questionnaire assessed liking for 139 foods and beverages using a 9-point hedonic scale. Genetic data was obtained from UK Biobank genotyping arrays, and imputation was performed to increase marker density. Quality control procedures were implemented to ensure data accuracy. After adjusting for covariates (age, sex, genetic principal components, array type, and batch), and accounting for genetic relatedness using GRAMMAR+, GWAS was performed using regscan with an additive model for SNPs with minor allele frequency (MAF) > 0.001. A study-wide significance threshold was determined based on the number of independent components explaining 95% of the variance across traits (p < 1.47 × 10⁻⁹). Hierarchical factor analysis was used to identify relationships between different food preferences, employing genetic correlations estimated using LD score regression. Genetic correlations between food liking and consumption (from the Pan UKBB project), and between food liking factors and other complex traits (from Idhub), were estimated. Locus definition was based on SNPs with p < 1 × 10⁻⁵, considering SNPs within 250kb as belonging to the same locus. HyPrColoc was used for co-localization analysis to identify true pleiotropy. Meta-analysis and replication were performed using 11 independent cohorts. Gene prioritization was conducted using haploR, prioritizing genes based on proximity to sentinel SNPs and the presence of non-synonymous or coding SNPs. Finally, a novel strategy was used to distinguish direct from mediated effects by modeling the effects of each SNP onto all nodes of the hierarchical food liking model using GenomicSEM. This methodology disentangled direct SNP-trait associations from those mediated through higher-order latent factors. Functional and tissue enrichment analyses were carried out using FUMA and clusterProfiler, respectively. Genetic correlation with brain MRI traits from the Oxford Brain Imaging Genetics Server was also performed using high-definition likelihood (HDL).

The GWAS identified three major dimensions of food liking: "Highly palatable" (high-reward foods), "Acquired" (foods with learned preferences), and "Low caloric" (vegetables, fruits, whole grains). The "Highly palatable" dimension was genetically independent of the others, suggesting distinct underlying mechanisms. A strong genetic correlation was observed between liking and consumption for most foods (r_g > 0.7), but liking consistently showed approximately twice the heritability compared to consumption (~0.08 vs. ~0.04), indicating a stronger biological influence on liking. Genetic correlations with other traits revealed expected patterns: "Highly palatable" liking associated with higher obesity indices, while "Low caloric" and "Acquired" likings showed the opposite. GWAS identified 1,401 significant SNP-trait associations across 173 loci (p < 1.47 x 10⁻⁹). Replication analysis confirmed the direction of effect for the vast majority of associations, despite lower replication rates attributable to lower power in the replication cohorts. Co-localization analysis using HyPrColoc indicated that most traits associated with the same locus also shared causal variants. A novel approach distinguished direct from mediated effects within the hierarchical model, identifying 495 direct effects. Gene prioritization revealed 250 genes, including taste and olfactory receptors, many with novel associations. Genes encoding bitter taste receptors were associated primarily with "Acquired" and "Low caloric" foods, while no such associations were found with "Highly palatable" foods. Genetic correlations with brain imaging data revealed that the "Highly palatable" dimension showed distinct associations with basal ganglia morphology (putamen and caudate), while "Acquired" and "Low caloric" dimensions showed associations with frontal, parietal, and occipital cortical thickness and surface area, as well as connectivity networks involving those areas.

This study's findings provide substantial evidence for a strong biological basis underlying food preferences, confirming and extending previous observations from twin studies. The distinct genetic architecture of the "Highly palatable" dimension, its independence from other dimensions, and its specific association with basal ganglia morphology suggest a unique neurobiological mechanism driving preference for high-reward foods, potentially involving dopaminergic pathways related to reward and motivation. The strong genetic correlation between liking and consumption, combined with the higher heritability of liking, supports the notion that liking is a fundamental driver of food choice, while environmental factors influence the transition from liking to actual consumption. The observed genetic correlations with other complex traits highlight the pleiotropic effects of genes influencing food preferences on overall health outcomes. The identification of numerous novel genetic associations further underscores the complexity of the genetic architecture of food liking and its potential connections with various physiological processes. The integration of genetic data with brain imaging data offers new insights into the neurobiological underpinnings of food choice.

This large-scale GWAS of food liking identified three distinct dimensions and 1,401 significant SNP-trait associations, many novel, highlighting the complexity of food preference genetics. The findings reveal distinct neurophysiological correlates for different food categories, emphasizing the biological basis of food liking and its influence on health outcomes. Future research should investigate the specific biological mechanisms underlying these genetic associations, including functional studies and examination of gene-environment interactions. Further, longitudinal studies will clarify the temporal stability of these associations and the influence of lifestyle on gene expression.

The study's reliance on self-reported data from the UK Biobank, a cohort known for selection bias toward healthier, more educated individuals, may limit the generalizability of the results. This bias may affect the interpretation of genetic correlations and the identification of certain loci. The cross-sectional nature of the data prevents a complete understanding of the causal relationships between food liking, consumption, and health outcomes. Future studies incorporating longitudinal data and more diverse populations are needed to address these limitations and enhance the generalizability of findings.