Genome-wide association studies of binge eating behaviour and anorexia nervosa yield insights into the unique and shared biology of eating disorder phenotypes

Abstract

Eating disorders -including anorexia nervosa (AN), bulimia nervosa, and binge eating disorder-are clinically distinct but exhibit symptom overlap and diagnostic crossover. Genomic analyses have mostly examined AN. We conducted the first genomic meta-analysis of binge eating behaviour (BE; 39,279 cases, 1,227,436 controls), alongside new analyses of AN (24,223 cases, 1,243,971 controls) and its subtypes (all European ancestries). We identified six loci associated with BE, including loci associated with higher body mass index (BMI) and impulse-control behaviours. AN GWAS yielded eight loci, validating six loci. Subsequent polygenic risk score analysis demonstrated an association with AN in two East Asian ancestry cohorts. BE and AN exhibited similar positive genetic correlations with psychiatric disorders, but opposing genetic correlations with anthropometric traits. Most of the genetic signal in BE and AN was not shared with BMI. We have extended eating disorder genomics beyond AN; future work will incorporate multiple diagnoses and global ancestries.

Figure 1:

Miami plots showing results from the BE-BROAD (top) and AN (bottom) meta-analyses. The dotted red line is the genome-wide significance threshold (P≤5×10⁻⁸). a. Main GWAS analyses, with variants reaching genome-wide significance coloured in blue if significant in the anorexia nervosa GWAS and in purple if significant in the binge eating broad GWAS. Variants reaching genome-wide significance in case-case GWAS of BE-BROAD vs AN are coloured in red. b. GWAS-by-subtraction analyses, showing results from the non-BMI genetic component. Variants reaching genome-wide significance coloured in blue for anorexia nervosa non-BMI component and purple in binge eating non-BMI component.

Figure 2:

Genetic distance between cases and controls of BE-BROAD and AN estimated by case-case GWAS. The genetic distance, $\sqrt{m\times F_{ST,\text{causal}}}$ is calculated by taking the square root of the product of $m$, the number of independent causal variants estimated here as 10,000 based on the polygenic nature of BE-BROAD and AN, and $F_{ST,causal}$, the average normalized squared differences in allele frequencies, derived based on the SNP-based heritabilities, genetic correlations, and population prevalences of the two traits

Figure 3:

Genetic correlations (r_g) of selected external traits with BE-BROAD and AN, split by those that differ significantly between the eating phenotypes (left) and those that do not (right). The r_gs were computed by Linkage Disequilibrium Score Regression (LDSC). The r_g estimates are indicated by the dots and standard errors are indicated by the lines on either side of each dot. r_g estimates have been corrected for multiple testing via the Bonferroni method. Information about the summary statistics used in our analysis can be found in Supplementary Table 8. BE-BROAD = binge eating broad definition; AN = anorexia nervosa; MDD = major depressive disorder; PGC = Psychiatric Genomics Consortium; BMI = Body mass index; F = female; M = male; FFM = fat-free mass; AUDIT-P = Alcohol Use Disorder Identification Test problem items