Population-based Risk of Psychiatric Disorders Associated with Recurrent CNVs

Abstract

Recurrent copy number variants (rCNVs) are associated with increased risk of neuropsychiatric disorders but their pathogenic population-level impact is unknown. We provide population-based estimates of rCNV-associated risk of neuropsychiatric disorders for 34 rCNVs in the iPSYCH2015 case-cohort sample (n=120,247). Most observed significant increases in rCNV-associated risk for ADHD, autism or schizophrenia were moderate (HR:1.42-5.00), and risk estimates were highly correlated across these disorders, the most notable exception being high autism-associated risk with Prader-Willi/Angelman Syndrome duplications (HR=20.8). No rCNV was associated with significant increase in depression risk. Also, rCNV-associated risk was positively correlated with locus size and gene constraint. Comparison with published rCNV studies suggests that prevalence of some rCNVs is higher, and risk of psychiatric disorders lower, than previously estimated. In an era where genetics is increasingly being clinically applied, our results highlight the importance of population-based risk estimates for genetics-based predictions.

Figure 1:. Population-based rCNV prevalence in iPSYCH2015 and comparison with UKB.

a) Population-based prevalence in iPSYCH2015 of deletions (x-axis) and duplications (y-axis) at 18 rCNV loci. Dashed line indicates del-dup prevalence parity. The 8 loci with significant del-dup prevalence disparity (P_FDR<0.05) are indicated on the plot. b) Population-based rCNV prevalence of deletions (left) and duplications (right) in iPSYCH2015 (full colour, above) compared to that reported for the UKB (semi-transparent colour, below), with loci ordered from top to bottom by decreasing sum LOEUF score. In the UKB study, 22q11.2b (turquoise) was not included as an independent locus and the prevalence of PWAS deletion was <1/100,000. Significant differences in prevalence between the two studies are indicated with asterisks (P_FDR<0.05 (*), <0.01 (**), and <0.001 (***), respectively). Error bars indicate the standard error (SE) of the weighted population-based prevalence in iPSYCH2015 and observed prevalence in UKB, respectively. For detailed results see Supplementary Table 1.

Figure 2:. rCNV-associated risk of psychiatric disorders in the iPSYCH2105 case-cohort.

rCNV-associated Hazard ratios (HR) and 95% confidence intervals (CI95%) were derived from Cox proportional hazards (CPH) models using inverse probability of sampling (IPS) weights, and are indicated for deletions and duplications by red and blue colour, respectively, for the 18 rCNV loci assessed in this study. Associated HRs with P<0.05 (i.e. with CI95% not overlapping HR=1) are bolded, and loci are ordered on the y-axis from top to bottom by decreasing sum LOEUF score. Abbreviations for the four most common psychiatric disorders targeted by the iPSYCH2015 case-cohort design (depicted on plot) are as follows: ADHD; attention-deficit hyperactivity disorder, ASD; autism spectrum disorder, MDD; major depressive disorder, SSD; schizophrenia spectrum disorder. Full per-locus results, including estimates also for Schizophrenia (SCZ), bipolar disorder (BPD), any affective disorder (AFF), any iPSYCH disorder (ANY), intellectual disability (ID) and epilepsy, are provided in Extended Data Figure 2.

Figure 3:. Contrasts in associated risk between main psychiatric outcomes across rCNVs.

a) Hazard ratios (HR) with error bars indicating standard errors (SE) for rCNV deletions (circle) and duplications (triangle) are plotted in a pairwise comparison for ADHD-v-ASD (left), SSD-v-ASD (middle), and SSD-v-ADHD (right), with estimates for each disorder on the x- and y-axis, respectively. The dashed line indicates risk parity. On each pairwise comparison plot, we have highlighted those rCNVs showing significant evidence of a rCNV-by-diagnosis interaction in generalized estimating equations (GEE) predicting case status for two diagnoses at a time (P_FDR<0.05), with colours corresponding to the rCNV locus in Figure 1. b) We constructed a heatmap of the coefficients from the rCNV-by-diagnosis interaction derived from GEE models for each pairwise comparison across all rCNVs. Significant coefficients are indicated with asterisks (P_FDR<0.05 (*), P_FDR<0.01 (**)). rCNV loci are ordered on the x-axis from left to right by decreasing sum LOEUF score, with coefficients for deletions and duplications shown in separate rows for each pairwise diagnosis comparison. Only rCNVs with available HRs for both compared diagnoses were included in each GEE analysis.

Figure 4:. Contrasts in associated risk by dosage change across rCNV loci.

Hazard ratios (HR) with error bars indicating standard errors (SE) for deletions and duplications are plotted on the x-axis and y-axis, respectively for ADHD, ASD and SSD. The 16 loci with available HR estimates for both deletions and duplications for at least one diagnosis are indicated by different colours (see figure legend) corresponding to previous figures. The dashed line indicates risk parity. We found no evidence of a locus-by-dosage interaction increasing prediction of case status in an omnibus test across all loci and diagnoses between a GEE model with the interaction term and a nested model without the interaction term (likelihood ratio test; P=0.16).

Figure 5:. rCNV-associated risk as a function of locus properties and rCNV prevalence.

Hazard ratios (HR) with error bars indicating standard errors (SE) for deletions (light red) and duplications (iris blue) associated with ADHD (left), ASD (middle), and SSD (right) are plotted against a) locus size and b) sum LOEUF score, as well as c) population prevalence in iPSYCH2015 and d) the population prevalence ratio between iPSYCH2015 and UKB. PWAS-dup was excluded because of outlying locus size and imprinting mechanism, and 13q12.12-dup was excluded from d) owing to the outlying low iPSYCH2015/UKB prevalence ratio. The plots include fitted trend lines (created with geom_smooth). The overall trend across all three psychiatric outcomes for the two locus features and the two CNV prevalence-related features was assessed with generalised estimating equation (GEE) models including all rCNV carriers (Methods), and significant positive associations were found for locus size (P=0.0028) and sum LOEUF score (P=0.0030) as well as the prevalence ratio between iPSYCH2015 and UKB (P=0.0086), but not CNV prevalence in iPSYCH2015 (P=0.31).