COLLABORATIVE STUDY OF THE COMBINED EFFECTS OF RARE CNVS AND POLYGENIC RISK ON PSYCHIATRIC TRAITS

Recurrent CNVs are known to be major risk factors in neuropsychiatric disorders, yet phenotypic variability between carriers of the same CNV remains unexplained. A possible explanation is that the effect of a CNV depends on genetic background, which can be quantified by a polygenic risk score (PRS). Using data from the PGC and UKBB (combined n=543,111), this project aims to characterize the individual and combined effects of recurrent CNVs and PRS on six major psychiatric disorders: ADHD, ASD, BD, PTSD, MDD, and SCZ.

We first quantified the effects of 42 recurrent CNV loci across all 6 disorders using logistic regression. After Bonferroni correction, 24 loci had a significant association with at least one disorder in either the deletion or duplication. We next documented evidence for the traditional liability threshold model of disease risk; cases carrying CNVs with weak main effects had higher PRS than cases carrying CNVs with strong main effects. This pattern was strongest for SCZ (p=2.13e-4) but was evident in BD as well. Additionally, we tested a composite CNV-PRS model, which demonstrates how PRS can be a useful tool for predicting outcomes in CNV carriers. For example, a carrier of 16p11.2 proximal duplication (a well-known SCZ association) is not at increased risk for SCZ if they have a low PRS-SCZ.

To increase power to detect statistical interactions between CNVs and PRS, we conducted a meta-analysis of CNVxPRS effects on BMI and height in four biobanks: UK Biobank, Estonian Biobank, Geisinger Health, and Million Veterans Program, (n=975,408). Of the 32 CNVs that were sufficiently powered for this analysis (n > 225), 3 had nominally significant (p < .05) interactions with PRS on BMI. In all three cases, the sign on the interaction was the same as the main effect of the CNV, suggesting that these interactions are synergistic. When we collapsed CNVs by their main effect direction, we saw a significant negative interaction between the BMI decreasing CNVs and PRS (p=9.98e-4). These interactions were robust to rescaling of the BMI response variable via inverse normalization or Box-Cox. We observed no significant interactions for Height.

Taken together, these analyses demonstrate that the effect of recurrent CNVs is moderated by PRS. In addition to emphasizing the importance of considering genetic background when studying the effects of rare variants, this study also demonstrates that genetic factors may have non-additive effects on complex traits.