After quality control of the GWAS data and removal of the related subjects, 1633 EAs and 2847 AAs were retained. We further removed 55 EAs and 81 AAs with unknown lifetime PTSD diagnosis. Final GWAS analyses included 1578 EAs (300 PTSD cases) and 2766 AAs (444 PTSD cases). Demographic information of the subjects is listed in .
Subject Demographic Information
In EAs and AAs, 768,146 and 870,103 SNPs, respectively, passed quality control steps. Based on Bonferroni correction for multiple comparisons, the threshold of genome-wide significance was 6.51 ×10−8 for EAs and 5.75×10−8 for AAs. The genomic inflation factor λs were 1.010 and 1.003 for EAs and AAs, respectively. The quartile-quartile plots are shown in . In EAs (), one SNP, rs406001 on chromosome 7p12, exceeded the genome-wide significant threshold for association with p = 3.97×10−8 (). Another two SNPs in the same region, rs382903 and rs450378, were the second and fourth most significant SNPs. The three SNPs are located approximately 630 kilobase downstream of the nearest gene Cordon-Bleu (COBL). The region showing the second strongest evidence of association was chromosome 4q32 (), which is in the first intron of Tolloid-Like 1 gene (TLL1). The top SNP rs6812849 reached a p value of 2.99×10−7. Another two SNPs in the same region, rs7691872 and rs1503292, were the sixth and seventh most significant SNPs. The SNPs ranked as the eighth and ninth most significant were located in the introns of NDRG1 and GABBR2, respectively, with p values of 3.1 × 10−6 and 5.4 × 10−6. Compared with EAs, GWAS results for AAs were less remarkable; no SNPs had p < 1 ×10−6 ().
Figure 1 Quintile-Quintile plots for whole cohort of European Americans (A), European Americans with trauma experiences (B), whole cohort of African Americans (C), and African Americans with trauma experiences (D). The genomic inflation factor (λ) is the (more ...)
Top Ten Most Significant SNPs in the GWAS Discovery Results for European Americans
Figure 2 Regional Manhattan plots of the top two strongest association regions in European Americans: chromosome 7p12 (A) and chromosome 4q32 (B). The single nucleotide polymorphisms are color coded based on the linkage disequilibrium with the most significant (more ...)
Top Ten Most Significant SNPs in the GWAS Results of the African Americans
Next, we restricted the GWAS analyses to PTSD control subjects with self-reported traumatic experiences. A total of 468 EAs and 933 AAs who did not report any traumatic events were excluded, leaving 1110 EAs and 1833 AAs for analyses. Distribution of different types of trauma exposure in EAs and in AAs is shown in Figure S1 in Supplement 1
. In both populations, witnessed someone being badly injured or killed was the most reported trauma experience, followed by seriously physically attacked or assaulted. In EAs, 32% reported one traumatic experience, 23% reported two traumatic experiences, and 45% reported three or more traumatic experiences. The distribution in AAs was very similar (Figure S2 in Supplement 1
). In EAs and AAs, 768,784 SNPs and 871,502 SNPs, respectively, passed quality control steps. The threshold of genome-wide significance was 6.50×10−8
for EAs and 5.74×10−8
for AAs. The genomic inflation factor λs were 1.017 and 1.007 for EAs and AAs, respectively. The quartile-quartile plots are shown in . Despite restricting the control subjects to trauma exposed subjects, which removed nearly one third of the sample, the GWAS results for EAs were very similar to those derived from the whole cohort (Figure S3 in Supplement 1
). Single nucleotide polymorphism rs406001 on chromosome 7p12 remained the most significant SNP, followed by rs382903 and rs450378 located in the same region of chromosome 7p12 with p
values ranked as the second and fourth in the results. The first intron of TLL1
remained the second most significant region. However, probably due to the reduced sample size, no SNPs reached genome-wide significance in the subsample. Genome-wide association study results in AAs also yielded no significant association regions.
We repeated the GWAS analyses again without removing the related subjects. Instead, we adjusted the genetic structure using factored spectrally transformed linear mixed models (25
). This left 1846 EAs (308 cases) and 3392 AAs (456 cases) for analysis after quality control steps. Apparently, due to the larger sample size, GWAS results for EAs were more significant than before (Table S1 in Supplement 1
). Single nucleotide polymorphism rs406001 on chromosome 7p12 remained the most significant SNP with p
. Single nucleotide polymorphisms rs382903 and rs450378 in the same region were the third and thirteenth most significant SNPs, and rs6812849 in the first intron of TLL1
was the second most significant SNP with p
= 1.21 × 10−7
. Another two SNPs in the first intron of TLL1
, rs1503292 and rs7691872, were the fourth and fifth most significant SNPs, respectively. After restricting the analysis to trauma-exposed subjects, the three SNPs on 7p12 were the first, fifth, and thirteenth most significant SNPs; the three SNPs in the first intron of TLL1
ranked as the sixth, seventh, and eighth most significant SNPs. Genome-wide association study results for AAs were also very similar (Table S2 in Supplement 1
). No SNPs reached genome-wide significance for AAs.
Based on the GWAS results from EAs, we TaqMan genotyped four SNPs in another 3112 subjects recruited using the same methodology for follow-up. These four SNPs were also genotyped in 5504 of the 5799 GWAS subjects for estimating genotyping error rates. Two of the four SNPs, rs406001 and rs450378, were located on chromosome 7p12. The linkage disequilibrium (R2) between them was .54. The other two SNPs, rs6812849 and rs7691872, were located in the first intron of TLL1. They were in high linkage disequilibrium (R2 = .88). By comparing the genotyping results from TaqMan and the GWAS array, the measured discordance rates for rs406001, rs450378, rs6812849, and rs7691872 were .20%, .65%, .71%, and .29%, respectively.
We excluded 85 of the 3112 replication samples due to an unknown PTSD diagnosis. In the remaining 3027 subjects, 1899 were EAs, including 207 PTSD cases. Of the four SNPs genotyped in the replication sample, the two located at TLL1, rs6812849 and rs7691872, were significantly associated with PTSD with p values of 6.3×10−6 and 2.3×10−4, respectively. Upon combining the GWAS and replication samples, rs6812849 reached genome-wide significance with p = 3.1×10−9, and rs7691872 reached p = 1.2×10−7. When the analyses were restricted to subjects with trauma exposure, the p values for rs6812849 and rs7691872 were 1.6×10−3 and 2.5×10−3, respectively, in the replication sample and 1.2×10−6 and 4.3×10−6, respectively, in the combined sample. The association between PTSD and the two SNPs on chromosome 7p12 was not replicated ().
In addition to chromosome 7p12 and chromosome 4q32, we followed up another four SNPs using a GoldenGate custom array (Illumina, San Diego, California) in 2553 replication subjects. After quality control, 1658 EAs (134 PTSD cases) and 744 AAs (89 PTSD cases) with known PTSD diagnosis were included in the replication analyses. The first SNP, rs2272651, was located in the seventh intron of NDRG1 and the second SNP, rs2779551, was located in the fourth intron of GABBR2. These two SNPs were the eighth and ninth most significant in the GWAS results for EAs and were selected because they map to obvious candidate loci. Neither of these SNPs was significantly associated with PTSD in the 1658 EA subject replication sample (). We also tested two SNPs with the most significant p values in the GWAS results of AAs: rs7014900, which is about 842 kilobase away from the nearest gene TRPS1, and rs13006863, which is located in the 14th intron of SLC4A5. Likewise, neither of them was replicated in the 744 AAs ().
Neither of the TLL1
SNPs, rs6812849 and rs7691872, was significantly associated with PTSD in AAs. Because AA is an admixed population with ~20% European ancestry (30
), we estimated the local ancestry around the two TLL1
SNPs of the GWAS AA subjects using HAPMIX (27
). Local estimated European ancestry of each subject was adjusted in the logistic regression model. Nevertheless, no significant association was observed (rs6812849, p
= .68; rs7691872, p
= .48). Among the 2766 AAs, 2266 had local African ancestry ≥50%. We restricted our analysis to these subjects, and no significant association was observed (rs6812849, p
= .95; rs7691872, p