Genome-wide association of bipolar disorder suggests an enrichment of replicable associations in regions near genes

Abstract

Although a highly heritable and disabling disease, bipolar disorder's (BD) genetic variants have been challenging to identify. We present new genotype data for 1,190 cases and 401 controls and perform a genome-wide association study including additional samples for a total of 2,191 cases and 1,434 controls. We do not detect genome-wide significant associations for individual loci; however, across all SNPs, we show an association between the power to detect effects calculated from a previous genome-wide association study and evidence for replication (P = 1.5×10(-7)). To demonstrate that this result is not likely to be a false positive, we analyze replication rates in a large meta-analysis of height and show that, in a large enough study, associations replicate as a function of power, approaching a linear relationship. Within BD, SNPs near exons exhibit a greater probability of replication, supporting an enrichment of reproducible associations near functional regions of genes. These results indicate that there is likely common genetic variation associated with BD near exons (±10 kb) that could be identified in larger studies and, further, provide a framework for assessing the potential for replication when combining results from multiple studies.

Figure 1. Replication as a function of power in height meta-analysis.

Power to detect associations in test data sets was calculated based on observed effects in discovery subsets of the IBC height collection and is plotted against association at P<0.05 in the test data. Lines are smoothed splines indicating the proportion of SNPs that replicate at P<0.05 across varying power. Smaller subsets of the discovery data set are shown in rainbow colors.

Figure 2. SNPs near exons show a stronger enrichment at P<0.05 as a function of power.

For different classes of SNPs, the smoothed spline is shown for the proportion of SNPs showing association at P<0.05 in the GAIN+TGEN dataset as a function of power based on the WTCCC dataset.