A multi-prs approach to predicting substance use disorders and associated comorbidities

Substance use disorders (SUDs) are a major public health concern worldwide and identifying genetic factors underlying SUDs can inform prevention and treatment strategies. Biobanks offer a valuable resource for identifying individuals with SUDs using electronic health records and predicting an individual's genetic risk for developing these SUDs using polygenic risk scores (PRS). A recent study combined genome-wide association studies (GWASs) of different SUDs to reveal a common genetic liability for addiction which predicts SUDs and their comorbidities. However, it remains unclear whether substance-specific PRSs add any predictive value beyond the addiction factor. We investigate this by leveraging access to large scale biobanks with data on SUDs and related psychopathology. We defined SUD cases and controls for alcohol (AUD), tobacco (TUD), cannabis (CUD), and opioid (OUD) use disorders using the presence of at least one diagnostic code. We similarly defined cases and controls for SUD-associated comorbidities such as MDD, PTSD, and bipolar disorder (BD). PRSs for the addiction factor, each SUD, and each comorbidity were estimated using summary statistics from the largest GWAS available. For the multi-PRS model, each SUD was modeled with logistic regression using all SUD-related PRSs as predictors while adjusting for age, sex, and ancestry. This model tests for association of each PRS with each SUD after accounting for the association of all other PRSs included in the model. Importantly, the model assesses whether substance-specific PRSs are informative beyond the addiction factor PRS. For the SUD comorbidities, the multi-PRS model also included the corresponding comorbidity PRS and assessed whether SUD-related PRSs are associated with the comorbidities after adjusting for the corresponding PRS. The results will assess these associations across multiple biobanks from the PsycheMERGE Consortium and across ancestries. Among 45000 participants in the Mayo Clinic Biobank with European ancestry, 963 (2.2%), 1738 (4.0%), 114 (0.3%), 161 (0.4%), 8701 (19.5%), 693 (1.6%), and 702 (1.6%) had a at least one diagnosis of AUD, TUD, CUD, OUD, MDD, BD, and PTSD, respectively. The addiction factor PRS was highly associated with all SUD diagnoses when considered in isolation. However, in multi-PRS models adjusting for SUD-specific PRSs, only AUD-PRS (OR = 1.32; p = 5e-8) and TUD-PRS (OR = 1.16; p = 9e-5) were associated with AUD, addiction factor PRS (OR = 1.20; p = 2e-5) and TUD-PRS (OR = 1.40; p = 5e-30) were associated with TUD, addiction factor PRS (OR = 1.55; p = 0.001) was associated with OUD, and no PRSs were significantly associated with CUD after accounting for multiple testing. In the multi-PRS models of SUD-associated comorbidities, the genetic liabilities for addiction and TUD were associated with increased odds for MDD and PTSD even after adjusting for the PRSs for MDD and PTSD, respectively. Addiction genetic liability was highly associated with all SUD diagnoses when considered in isolation, however, substance-specific PRSs were more associated in multi-PRS models, suggesting that substance-specific PRSs have some predictive value beyond the addiction factor. The SUD-related PRSs were also associated with MDD and PTSD after adjusting for genetic liability for each disorder. These findings have important implications for developing prevention and treatment strategies for SUDs and their associated comorbidities.