Single Marker Analysis
In order to replicate and assess the specificity of 14 selected genes from published GWAS on BPD, we analyzed 88 tag SNPs in the genes ANK3, ARNT2, CACNA1C, CMTM8, DFNB31, DGKH, EGFR, NALCN, NPAS3, NXN, SLC19A3, SLC29A3, SLC39A3, and SORCS2 in our BPD sample. After correction for multiple testing, two SNPs (rs1170169 and rs9525580) in DGKH remained significantly associated with BPD; at the nominal level, eight further significant findings were detected in the combined BPD sample, while another 13 SNPs were associated in only one of the BPD subsamples (; Supplementary Table 1). In order to examine the specificity of these associations for BPD, we further analyzed all 23 nominally associated markers (from the genes CMTM8, EGFR, DFNB31, DGKH, NPAS3, and SLC39A3) in UPD (; Supplementary Table 3) and aADHD (; Supplementary Table 5). SNPs from the other eight genes showed no significant association with BPD and were thus not analyzed further.
Cross-disorder genotyping revealed a total of 12 SNPs in four genes (CMTM8, DGKH, NPAS3, and SLC39A3) that were associated with at least one of the three examined phenotypes at the nominal level (). Ten association p-values were nominally significant in the combined BPD sample (the remaining other two had a borderline significant p=0.056), seven in the combined UPD sample, and six in the aADHD sample, with seven SNPs being associated with two and two SNPs with all three disorders. Only those SNPs that were associated with BPD in the combined sample replicated in either UPD or aADHD (however, including rs2148004 and rs347405 with p=0.056), but not those that were only associated in one of the subsamples (compare Supplementary Table 1 with ). Most of the replicated SNPs mapped to DGKH (BPD: six SNPs; UPD: seven SNPs; aADHD: four SNPs), which was the only gene in our study that contained SNPs, which were significant following Bonferroni correction (). Noteworthy, all associations found with UPD mapped to DGKH. In CMTM8, only one SNP (rs6803740) featured an overlapping association between BPD and aADHD. The same disorders also overlapped in their association regarding the NPAS3 SNP rs7455703. The SLC39A3 association of rs4806874 was found to be exclusive for BPD ().
Haplotype analysis was then performed with all cross-disorder genotyped SNPs (BPD, significant findings: , complete data are given in Supplementary Table 2; UPD significant findings: , complete data are given in Supplementary Table 4; aADHD significant findings: , complete data are given in Supplementary Table 6). The strongest association found in all analyzed disorders was in DGKH block 2 (rs994856–rs9525580–rs9525584; ) haplotype GAT, which is exclusively composed of each single marker's risk alleles; this was consistent in all three examined phenotypes. Accordingly, GAT frequency was increased in all case groups as compared with controls; although this was nominally significant in all three disorders, the FWER was below 5% only in BPD and aADHD, but slightly above this threshold in UPD (permutation p=0.056). The GAT haplotype can, therefore, be assumed to predispose to at least two, but possibly to any of the three disorders (see ). In terms of frequency, GAT follows its ‘complementary' haplotype AGC, which is composed of those alleles that have a higher MAF in controls. The expected protective effect conveyed by AGC, however, was only significant in UPD following FWER correction, and nominally also in BPD (). A similar phenomenon was seen in DGKH block 1 (rs1170191–1170169–rs2148004) haplotype GCG, whose frequency was lower in all case groups as compared with controls, but the presumed protective effect was nominally significant only in BPD and UPD. The haplotype GGA in turn was enriched in all cases; following FWER, this was significant in BPD, whereas nominally it was also associated with aADHD and UPD, respectively (p=0.051; ). Two further risk haplotypes were exclusively found to be associated with UPD ().
Association of Haplotypes With the Complete BPD, UPD, and aADHD Samples Containing At Least One Significant Haplotype
LD plot of DGKH, according to the four-gamete rule.
Haplotype associations in genes other than DGKH were found to be restricted to specific psychiatric disorders and did not withstand correction for multiple testing. In CMTM8, block 1 (rs6550109–rs12496256; ) haplotype TG was significantly protective, while block 3 (rs4276227–rs6803740) haplotype CG was associated with risk for BPD (). The CMTM8 block 2 (rs4955272–rs7644602–rs7632109) haplotype GGG was the only significant haplotype association in aADHD and presumed to be protective (). NPAS3 rs8015959–rs17455703 had two alleles associated with BPD, the protective CG and the risk haplotype CA ().
LD plot of CMTM8, according to the four-gamete rule.
To compare our findings with previous studies, we have subjected the significant SNPs rs9315885 and rs1170191 from the studies by Baum et al (2008a)
, Ollila et al (2009)
, and Squassina et al (2009)
to a formal meta-analysis (; Supplementary Table 8). While rs9315885, which was significant in the three other studies but not our data set, proved to be highly significantly associated in the meta-analysis (), this was not the case for rs1170191, as the effect direction was reversed in our as compared with other studies ().
Forest plots displaying meta-analyses of minor vs major allele of rs9315885 (a) and rs1170191 (b).