Genetic clusters of BMI reveal symptom-specific causal effects on depression.
While BMI is phenotypically and genetically associated with depression, the extent to which BMI causes depression and the mechanisms underlying this effect remain unclear. We estimated the causal effect of BMI on depression symptoms, accounting for complexity in both the exposure and the outcome. We applied PheWAS-based Clustering of Mendelian Randomization instruments (PWC-MR) to partition 324 BMI-associated SNPs into six genetic clusters and estimated the effect of each cluster on nine depression symptoms from the largest available genome-wide meta-analysis of PHQ-9 depression symptoms (N = 224,535-308,421). Across all six clusters, BMI had the largest effect on appetite changes (β ranges from 0.19 to 0.34), which was robust across estimators, homogeneous across clusters, and survived correction for multiple testing. Smaller effects were observed for several other symptoms, including concentration changes, fatigue, anhedonia, and sleep problems, but these were heterogeneous across clusters and often attenuated under pleiotropy-robust estimators, indicating likely bias from horizontal pleiotropy. Little evidence of an effect was detected for depressed mood and suicidal ideation. Cluster-level heterogeneity was observed for the majority of symptoms, demonstrating that the causal effect of BMI varies by instrument group and indicating likely violation of the exclusion restriction assumption. These findings indicate that the causal effect of BMI on depression detected in typical non-clustered Mendelian randomization is driven to an extent by appetite, with no or inconsistent evidence for effects on core psychological symptoms such as anhedonia and depressed mood.