Deep resequencing and association analysis of schizophrenia candidate genes

In 2005, we selected 10 genes for which there was reasonable evidence for involvement in the etiology of schizophrenia (COMT, DAOA, DISC1, DRD2, DRD3, DTNBP1, HTR2A, NRG1, SLC6A3 and SLC6A4, Supplementary Table S1).¹ Although these genes have not received support from far larger and comprehensive subsequent studies, and may not contain common etiological variations,² it is possible that they contain uncommon variations of etiological importance. To test this hypothesis, we conducted a multistage resequencing study.

In stage 1, we used Sanger methods to sequence the exons, 5′ and 3′ untranslated regions, splice sites, promoters and conserved intronic regions of these 10 genes in 727 cases with schizophrenia from CATIE³ and 733 controls of European (EUR) and African (AFR) ancestry. In stage 2, we validated single-nucleotide variants (SNVs) using Roche 454 sequencing in the same samples. In stage 3, we genotyped prioritized SNVs in independent samples (Supplementary Material and Supplementary Figure S1).

In stage 1, Sanger sequencing identified 782 variants, including 587 novel SNVs that are not found in dbSNP132 (Supplementary Tables S2–S4). As expected, the number of novel variants discovered per individual was higher in those of AFR (1.46) than EUR ancestry (0.95), but cases and controls did not differ (EUR cases/controls: 0.920/0.980; AFR cases/controls: 1.492/1.430). The numbers of SNVs per gene were also similar, although DISC1 showed a nonsignificant excess in cases (EUR cases/controls: 0.138/0.119; AFR cases/controls: 0.243/0.167) mostly owing to novel variants in AFR subjects (cases/controls: 0.173/0.099). Three unrelated cases, but zero controls, were each found to have a single novel nonsense mutation: two for DISC1 (truncating only the ‘Es’ splice variants) and one for SLC6A4 (Supplementary Tables S2–S4). The ratio of nonsynonymous to synonymous variants was similar in cases and controls (1.25 vs 1.16).

In stage 2, we prioritized 254 of the 782 variants for technical replication, as they met at least one of the following criteria: (1) novel nonsense, missense or splice site variant, (2) novel intronic variant in ⩾1 EUR case, (3) novel variant with an odds ratio >2 in the EUR cohort and (4) dbSNP nonsense, missense or splice site variant in ⩾1 EUR case. Validation sequencing by Roche 454 revealed 225 true variants and 29 false positives (accuracy rate of 89%; Supplementary Figure S2).

In stage 3, we selected 92 out of the 225 SNVs (Supplementary Table S5) from stage 2 for genotyping in an independent sample of 2191 cases and 2659 controls (EUR and AFR). We included: (1) all novel nonsense, missense or splice site variants seen in ⩾1 case, (2) all variants seen in >1 case and (3) three nonsense variants observed in one case each. After genotyping 92 SNVs in the replication samples, 29 were monomorphic (22 of these were seen in only one case in stage 1), 6 had low-quality genotypes and 57 SNVs were tested for association with schizophrenia (logistic regression, separately for EUR and AFR subjects). Table 1 lists the SNVs with the smallest P value in each gene (complete results in Supplementary Table S6). No gene contained a SNV reaching the criteria for genome-wide significance (P<5 × 10⁻⁸). We then tested the aggregate effects of uncommon variants within a gene for 35 non-intronic SNVs with minor allele frequency (MAF) <0.01 (Table 1). No gene was significant following correction for multiple testing. For example, of the 20 uncommon variants in DISC1 (Supplementary Figure S3), there were 32 minor alleles in EUR cases and 40 in EUR controls.

Table 1 Summary of stage 3 replication genotyping results

Thus, multistage resequencing of 10 schizophrenia candidate genes did not yield support for uncommon exonic variation. This result is consistent with common variation results that do not, to date, provide support for these genes despite a sample size of 21 856 individuals. The replication sample had 80% power to detect a genotypic relative risk of 3.2 in the EUR cohort and 5.1 in the AFR cohort, with an MAF of 0.01 and significance level of P=5 × 10⁻⁸. For a relaxed threshold (P=0.001), there would have been >99% power to detect a genotypic relative risk of 2.9 in the EUR cohort and 4.6 in the AFR cohort.

The stage 1 and 2 results hinted that DISC1 might contain an excess of uncommon variants, and DISC1 was thus the main focus in stage 3 replication. However, there was no evidence in our results to support the hypothesis that DISC1 contains uncommon variants that are of relevance to schizophrenia, as no single-SNV or aggregate test approach was even of nominal significance. Although DISC1 nonsense variants were present in 2 stage 1 cases and 0 controls, no additional cases or controls in the larger stage 3 sample had those particular nonsense mutations.

DISC1 has been the focus of dozens of genetic studies.⁴ However, a recent comprehensive meta-analysis of common variation did not support its role in schizophrenia susceptibility.⁵ To our knowledge, four groups have sequenced DISC1 in cases with schizophrenia,^{6, 7, 8, 9} and all had discovery samples far smaller than reported here (34, 90, 198 and 288 cases). Of the association results in these studies, none met the criteria for genome-wide significance. Only one study had a replication component, and the initial finding did not replicate.⁹ Therefore, despite employing a replication sample three times the size of the discovery sample, DISC1 was not found to contain common or uncommon variants individually, or in aggregate associated with schizophrenia.

The results of this study suggest that classical schizophrenia candidate genes do not harbor uncommon coding region variations of etiological importance.