Genomics yields biological and phenotypic insights into bipolar disorder

Abstract

Bipolar disorder is a leading contributor to the global burden of disease¹. Despite high heritability (60-80%), the majority of the underlying genetic determinants remain unknown². We analysed data from participants of European, East Asian, African American and Latino ancestries (n = 158,036 cases with bipolar disorder, 2.8 million controls), combining clinical, community and self-reported samples. We identified 298 genome-wide significant loci in the multi-ancestry meta-analysis, a fourfold increase over previous findings³, and identified an ancestry-specific association in the East Asian cohort. Integrating results from fine-mapping and other variant-to-gene mapping approaches identified 36 credible genes in the aetiology of bipolar disorder. Genes prioritized through fine-mapping were enriched for ultra-rare damaging missense and protein-truncating variations in cases with bipolar disorder⁴, highlighting convergence of common and rare variant signals. We report differences in the genetic architecture of bipolar disorder depending on the source of patient ascertainment and on bipolar disorder subtype (type I or type II). Several analyses implicate specific cell types in the pathophysiology of bipolar disorder, including GABAergic interneurons and medium spiny neurons. Together, these analyses provide additional insights into the genetic architecture and biological underpinnings of bipolar disorder.

Extended Figure 1.. Network diagram of the genetic correlations between BD ascertained from Clinical, Community and Self-report samples, as well as BD-subtypes (BDI and BDII).

The line widths are proportional to the strength of the correlations between pairs. BDI: bipolar disorder I, BDII: bipolar disorder II

Extended Figure 2:. Univariate MiXeR estimates of the required effective sample size needed to capture 50% of the genetic variance (horizontal dashed line) associated with each BD ascertainment and subtype.

N and Sample size refer to the effective sample size. The estimated effective sample size (and standard errors) are given in the legend alongside each trait name.

Extended Figure 3.. Trivariate MiXeR estimates for the genetic overlap of BD from Clinical, Community and Self-report samples.

The percentages show the proportion of trait-influencing variants within each section of the Venn diagram relative to the sum of all trait-influencing variants across all samples. The size of the circles reflects the polygenicity of each trait.

Extended Figure 4.. Miami plot for BD genome-wide meta-analyses, including all cohorts.

Upper panel: the multi-ancestry meta-analysis identified 298 genome-wide significant (GWS) loci. Lower panel: porcupine plot showing the results of the Latino (0 GWS loci), African American (0 GWS loci), East Asian (1 GWS locus) and European (229 GWS loci) meta-analyses. The x-axes show genomic position (chromosomes 1–22), and the y axes show statistical significance as –log10[p-value]. P-values are two-sided and based on an inverse-variance-weighted fixed-effects meta-analysis. The dashed black lines show the GWS threshold (P < 5 × 10⁻⁸). The star indicates the position of the East Asian GWS locus (rs117130410, 4:105734758, build GRCh37).

Extended Figure 5.. Cluster-level SNP-heritability enrichment for bipolar disorder.

The t-distributed stochastic neighbor embedding (tSNE) plot (left) (from Siletti et al.) is coloured by the enrichment z-score. Grey indicates non-significantly enriched superclusters (FDR > 0.05). The barplot (right) shows the top 35 significantly enriched clusters. The numbers in parentheses on the y-axis indicate the cell type clusters as defined in Siletti et al.

Extended Figure 6.. Number of SNPs within the smallest 95% credible sets (CS) from meta-analysis of European and multi-ancestry meta-analyses when excluding and including self-report data.

Colours represent CS of varying size, with blue CS containing 0 SNPs and red CS containing 15+ SNPs. All fine-mapped SNPs regardless of their PIP were used to assess the size of the 95% credible sets.

Extended Figure 7.. Methods and criteria for credible gene identification.

Extended Figure 8.. Clustering of patterns of temporal variation in expression of 34 credible genes.

Cluster 1 (n=21 genes) shows reduced prenatal gene expression, with gene expression peaking at birth and remaining stable over the life-course. Cluster 2 (n=13 genes) includes genes with a peak gene expression during fetal development with a drop-off in expression before birth. Genes within each cluster are described in Supplementary Table 31. To illustrate the variability in gene expression within each cluster we plot each donor expression value in each sampled brain region for the 34 credible genes as individual points. Smoothing splines used to illustrate the age trajectory for each cluster is based on generalized additive models with the predicted 95% confidence interval in grey. We use age in days to plot the variation in gene expression with the x-axis on a log2 scale and labels for birth, 10, 18, and 65 years of age as reference points.

Figure 1.

Genetic correlation and bivariate MiXeR estimates for the genetic overlap of BD ascertainment and subtypes. Trait-influencing genetic variants shared between each pair (grey) and unique to each trait (colours) are shown. The numbers within the Venn diagrams indicate the estimated number of trait-influencing variants (and standard errors) (in thousands) that explain 90% of SNP heritability in each phenotype. The size of the circles reflects the polygenicity of each trait, with larger circles corresponding to greater polygenicity. The estimated genetic correlation (rg) and standard error between BD and each trait of interest from LDSC is shown below the corresponding Venn diagram. Clinical and Community samples were stratified into bipolar I disorder (BDI) and bipolar II disorder (BDII) subtypes if subtype data were available. Model fit statistics indicated that MiXeR-modelled overlap for bivariate comparisons including the bipolar subtypes (BDI and BDII) were not distinguishable from minimal or maximal possible overlap, and therefore to be interpreted with caution (see Supplementary Table 4).

Figure 2.

Genetic correlations (with standard errors) between bipolar disorder and other psychiatric disorders. The y-axis (Trait2) is ordered based on the significance and magnitude of genetic correlation of each trait with bipolar disorder type I. P-values were calculated from the two-sided z statistics computed by dividing the estimated genetic correlation by the estimated standard error, without adjustment. The standard error for a genetic correlation was estimated using a ratio block jackknife over 200 blocks. Triangles indicate significant results passing the Bonferroni corrected significance threshold of two-sided P < 3.6 × 10⁻⁵. Each error bar represents the standard error of the estimate. ADHD, attention deficit/hyperactivity disorder. PTSD, post-traumatic stress disorder. The year indicated in parentheses after each trait refers to the year in which the GWAS was published. Details are provided in Supplementary Table 13.

Figure 3.

Phenotypic variance in bipolar disorder in European cohorts explained by polygenic risk scores derived from the multi-ancestry and European meta-analyses (with and without self-reported data). Variance explained is presented on the liability scale, assuming a 2% population prevalence of bipolar disorder. The results in the first panel (A) are the median weighted liability R² values across all 55 European cohorts (40,992 cases, 80,215 controls, N_eff = 46,725). Similarly, the remaining panels show the results across (B) 36 bipolar disorder I (BDI) cohorts (12,419 cases and 33,148 controls, N_eff = 14,607), (C) 21 bipolar disorder II (BDII cohorts, 2,549 cases, 23,385 controls, N_eff = 4,021), (D) 48 Clinical cohorts (27,833 cases, 46,623 controls, N_eff = 29,543), and (E) 7 Community cohorts (13,159 cases, 36,592 controls, N_eff = 17,178). All analyses were weighted by the effective n per cohort. The median liability R² is represented as a horizontal black line.

Figure 4.

Supercluster-level SNP-heritability enrichment for bipolar disorder. The t-distributed stochastic neighbour embedding (tSNE) plot (from Siletti et al.) (left) is coloured by the enrichment z-score. Grey indicates non-significantly enriched superclusters (FDR > 0.05). The barplot (right) shows the nine significantly enriched superclusters.