Phenotype integration improves power and preserves specificity in biobank-based genetic studies of MDD

Abstract

Biobanks often contain several phenotypes relevant to a given disorder, and researchers face complex tradeoffs between shallow phenotypes (high sample size, low specificity and sensitivity) and deep phenotypes (low sample size, high specificity and sensitivity). Here, we study an extreme case: Major Depressive Disorder (MDD) in UK Biobank. Previous studies found that shallow and deep MDD phenotypes have qualitatively distinct genetic architectures, but it remains unclear which are optimal for scientific study or clinical prediction. We propose a new framework to get the best of both worlds by integrating together information across hundreds of MDD-relevant phenotypes. First, we use phenotype imputation to increase sample size for the deepest available MDD phenotype, which dramatically improves GWAS power (increases #loci ~10 fold) and PRS accuracy (increases R2 ~2 fold). Further, we show the genetic architecture of the imputed phenotype remains specific to MDD using genetic correlation, PRS prediction in external clinical cohorts, and a novel PRS-based pleiotropy metric. We also develop a complementary approach to improve specificity of GWAS on shallow MDD phenotypes by adjusting for phenome-wide PCs. Finally, we study phenotype integration at the level of GWAS summary statistics, which can increase GWAS and PRS power but introduces non-MDD-specific signals. Our work provides a simple and scalable recipe to improve genetic studies in large biobanks by combining the sample size of shallow phenotypes with the sensitivity and specificity of deep phenotypes.

The genetic basis of major depressive disorder

“…We then imputed Lifetime MDD in the entire cohort, using 216 other phenotypes in the biobank, chosen regardless of their putative relationship with MDD, using Soft-Impute [134] (a variant of principal component analysis that accommodates missing data, and uses observed phenotype data to identify latent factors). We were able to show that the top phenome-wide factors capture pleiotropic axes for MDD, allowing us to identify genetic effects that are specific to lifetime MDD (which stood in for the gold standard MDD cases) [126]. Remarkably, we found that the one-item self-assessment measures (which capture general dysphoria), residualized of these latent factors, index core, MDD-specific biology.…”

“…We need a set of well phenotyped cases to seed imputation, but how to maximize imputation's effectiveness remains an open question, so it's not possible to provide robust estimates of the number of interview-based cases required. As an example of what is possible, imputation using data from a questionnaire-based measure of MDD from 67,164 UK biobank into 337,126 individuals with a single-item measure increased both the number of risk loci identified and out-of-sample prediction of MDD accuracy, while preserving better specificity to MDD than the single-item measure [126].…”

“…Many phenotypes in biobanks correlate with MDD, and these can be used as proxies for information we are missing. As a demonstration of this we analyzed the UK Biobank, taking the CIDI-SF based Lifetime MDD phenotype to represent a goldstandard assessment [126]. We then imputed Lifetime MDD in the entire cohort, using 216 other phenotypes in the biobank, chosen regardless of their putative relationship with MDD, using Soft-Impute [134] (a variant of principal component analysis that accommodates missing data, and uses observed phenotype data to identify latent factors).…”

Polygenic analyses show important differences between MDD symptoms collected using PHQ9 and CIDI-SF

“…As MR results show that PHQ9 symptoms are more likely endorsed due to non-MDD phenotypes that index general dysphoria than MDD, we further ask if this means that genetic studies on PHQ9 symptoms are less likely to lead us to identify MDD-specific biology. To do this we computed 10-fold cross-validated PRS on all PHQ9 and WorstEpisode symptoms in UKBiobank, and used these to predict MDD and non-MDD phenotypes explored above, obtaining the ratio between their prediction R 2 (PRS Pleiotropy = R 2 non-MDD /R 2 MDD ) 33 . A higher PRS Pleiotropy means lower specificity of a PRS for MDD.…”

“…For all quantitative phenotypes, including neuroticism, we evaluated accuracy using ordinary R 2 . PRS Pleiotropy 33 is calculated for each PHQ9 and WorstEpisode symptom using the ratio of its PRS predictions on 50 non-MDD phenotypes and its prediction on LifetimeMDD in UKBiobank or ICD10-based MDD in iPSYCH cohorts (PRSPleiotropy = R 2 non-MDD / R 2 MDD , Supplementary Tables 10-12 ).…”