FMR1, circadian genes and depression: suggestive associations or false discovery?

Abstract

Background: There are several indications that malfunctions of the circadian clock contribute to depression. To search for particular circadian gene polymorphisms associated with depression, diverse polymorphisms were genotyped in two samples covering a range of depressed volunteers and participants with normal mood.

Methods: Depression mood self-ratings and DNA were collected independently from a sample of patients presenting to a sleep disorders center (1086 of European origin) and from a separate sample consisting of 399 participants claiming delayed sleep phase symptoms and 406 partly-matched controls. A custom Illumina Golden Gate array of 768 selected single nucleotide polymorphisms (SNPs) was assayed in both samples, supplemented by additional SNPlex and Taqman assays, including assay of 41 ancestry-associated markers (AIMs) to control stratification.

Results: In the Sleep Clinic sample, these assays yielded Bonferroni-significant association with depressed mood in three linked SNPs of the gene FMR1: rs25702 (nominal P=1.77E-05), rs25714 (P=1.83E-05), and rs28900 (P=5.24E-05). This FMR1 association was supported by 8 SNPs with nominal significance and a nominally-significant gene-wise set test. There was no association of depressed mood with FMR1 in the delayed sleep phase case-control sample or in downloaded GWAS data from the GenRED 2 sample contrasting an early-onset recurrent depression sample with controls. No replication was located in other GWAS studies of depression. Our data did weakly replicate a previously-reported association of depression with PPARGC1B rs7732671 (P=0.0235). Suggestive associations not meeting strict criteria for multiple testing and replication were found with GSK3B, NPAS2, RORA, PER3, CRY1, MTNR1A and NR1D1. Notably, 16 SNPs nominally associated with depressed mood (14 in GSK3B) were also nominally associated with delayed sleep phase syndrome (P=3E10-6).

Conclusions: Considering the inconsistencies between samples and the likelihood that the significant three FMR1 SNPs might be linked to complex polymorphisms more functionally related to depression, large gene resequencing studies may be needed to clarify the import for depression of these circadian genes.

Figure 1

Minor allele frequency versus QIDS score. For the Sleep Clinic sample, the rs25702 minor allele frequency (MAF) with its 95% confidence interval range is plotted versus QIDS score ranges. QIDS scores correspond to the following levels of depression: mild (6–10), moderate (11–15), severe (16–20), and very severe (≥21) depression, where patients with scores <6 are not likely depressed and those >20 would usually not be observed in the clinic because they would need hospitalization. The lower the QIDS and the less depression observed, the more likely a patient was to have a minor allele.

Figure 2

QIDS scores for 0, 1, or 2 minor alleles. For the Sleep Clinic sample, the QIDS depression score is plotted (with 95% confidence limits) for zero, one, or two minor alleles in rs25702. Since the sample was largely male, there were only 3 females (less than 1%) with two minor alleles in rs28900, rs25714, or rs25702. The small number with two minor alleles accounted for the large confidence limit in QIDS scores for that group. It appeared that women with one or two minor alleles had averaged about the same QIDS score, suggesting a dominant trait associated with lower QIDS score.

Figure 3

Q-Q plot for Sleep Clinic SNPs. The ranked observed and randomly-expected linear regression probabilities (P values) are plotted for the Sleep Clinic SNPs on a negative log₁₀ scale. Higher values indicate smaller P values with greater significance. The three upper SNPs (from left to right) are FMR1 rs28900, rs25714, and rs25702. The red line indicates the expected trend were there no genomic inflation.

Figure 4

Stratified Q-Q plot for SNPs Associated with QIDS-SR. For SNP association with QIDS-SR, the ranked observed meta-analysis probabilities (P values) are plotted versus the expected probabilities. The scale is –log₁₀(P), so that points above and to the right represent more significant (i.e. smaller) P values. Unlike Figure 3, only the 614 SNPs are plotted for which meta-analysis P values were available both for the QIDS-SR (Additional file 1, worksheet 1) and for DSPS classification in a dominant logistic genetic model. The P values for QIDS-SR association are stratified by whether the association of that SNP with DSPS was nominally significant (P<0.05). The expectation for 614 SNPs with random association is plotted in the black dashed line. Since the distance of the most deviant p values for P_DSPS<0.05 (red line) is approximately 1 log unit from the random expectation (black dotted line) and from the line for P_DSPS>0.05 (blue line), one may estimate that the probability of false discovery of pleiotropy for the most significant SNPs was ~0.10. FMR1 SNPs were not included in Figure 4 because the dominant model could not be computed for this part of the X chromosome, but no FMR1 SNP was significantly associated with DSPS in an additive model.