EXPLORING THE IMPACT OF INCLUSION/EXCLUSION CRITERIA ON THE GENETIC ARCHITECTURE OF MAJOR DEPRESSIVE DISORDER IN DANISH BIOBANKS

Large samples are needed for genome-wide association studies (GWAS) of Major Depressive Disorder (MDD), and, indeed, recent studies aggregate more than a million individuals. Achieving these numbers means relaxing inclusion criteria for cases, often to a single self-report item, which may impact our ability to identify core disease mechanisms. Prior work suggests “shallow” criteria offer a different picture of the genetic architecture of MDD than more careful clinical criteria. Danish health registers, with long follow-ups, provide life-course medical history for those diagnosed with MDD and an ability to ask how representative samples are of the underlying population. Here, we study 50,000 MDD cases and 300,000 controls from three Danish biobanks with linkage to National register data: the iPSYCH2015 case-cohort study, the Danish Blood Donors Study, and the Copenhagen Hospital Biobank. We conduct a systemic analysis of the impact of ascertainment and inclusion/exclusion criteria on the inferred genetic architecture of MDD, seeking to clarify analogues for deeply phenotyped MDD. Among broadest possible MDD (BP-MDD; at least one diagnosis of ICD10: F32-33), we assessed the prevalence of potentially exclusionary diagnoses that reflect enrollment requirements for deeply phenotyped clinical studies (e.g., lifetime exclusions for schizophrenia, bipolar, or intellectual disability; post exclusions for onset of MDD subsequent to dementia or terminal illness diagnosis; event-based exclusions of MDD occurring within one year of an AUD, DUD or MCI diagnosis; and age-based exclusions for onset before 18 or after 50). We compare the genetic architecture of BP-MDD to clinically plausible MDD defined by the above criteria with or without age restrictions (CPA-MDD / CP-MDD) using GWAS and SNP-based variance components analysis. We then use polygenic score (PGS) profiles to test for differences in underlying genetic liability among cases with and without each potential exclusion criterion. Next, we compare the replication sensitivity of 250 index SNPs of previously identified MDD loci to inclusion/exclusion criteria. We observed an impact of exclusion criteria on GWAS power, replication sensitivity, and PGS profiles of individuals diagnosed with MDD. For example, we observed the PGS for MDD trended higher in CP-MDD while the PGS for SCZ and PTSD were significantly lower. Consistent with this, for some variants, effect sizes appear to increase when replicated in the smaller, more strictly defined CP-MDD, while other variants saw effect sizes diminish when moving from shallower BP-MDD to stricter CP-MDD. Our study highlights the impact that considering life-course exclusion criteria in defining MDD cases from biobanks has on the underlying genetic architecture. We believe this motivates important discussions regarding the next steps for MDD GWAS that could help improve the portability of results across cohorts and enable.