The idea that circadian rhythms had some role in affective disorders arose from clinical observations of their altered sleep-wake cycles, the cyclicity of the symptoms, and early work by pioneering researchers 1234. Attempts to identify circadian abnormalities in depressed and bipolar patients by physiologic means have yielded somewhat inconsistent and disappointing results 567. However, the now-proven efficacy of bright light treatment as well as a broader range of effective interventions in the circadian system provide strong evidence that circadian rhythms are somehow involved in the pathophysiology of affective disorders 8.

The effects of light treatment, along with the symptom development of seasonal affective disorder (often a bipolar phenotype), might suggest that mechanisms which trigger mood swings in humans resemble the circadian-controlled photoperiodic mechanisms governing mammalian seasonality 9. Recently, a number of studies of nocturnal and diurnal rodents have demonstrated influences of photoperiod upon animal models of depression 101112. Several reports have presented rationales and preliminary suggestive evidence that circadian system genetic abnormalities might contribute to affective disorders 131415161718. In addition, accumulating evidence indicates that heritable circadian disorders such as delayed sleep phase disorder are comorbid with depression 19. This may suggest that there are genetic polymorphisms in the circadian system which confer susceptibility both to depression and to delayed sleep phase disorder or its converse, advanced sleep phase disorder.

Genome-wide association studies of bipolar disorder have given no substantial support for a role of the circadian system 2021, although in one study, VGCNL1, a gene which may have a circadian role, came close to genome-wide significance 22. Genome-wide studies, however, are designed to detect common allelic variants of small effect, and do not exclude other types of gene effects, such as rare variants of strong effect.

Though the genome-wide association method may eventually replace the testing of candidate genes, we have thought it worthwhile to survey likely single nucleotide polymorphisms (SNPs) in the set of genes which form the circadian system through complex interactions. Most of the SNPs we have considered have not been tested directly in whole-genome association studies. Moreover, in some models, we have used transmission disequilibrium tests (TDT) with parent-proband trios or affected sib pairs which eliminate population stratification as a potential source of false-negative results. Here we report results of 4 ongoing studies which provide some cross-replication, and taken together, suggest that several circadian polymorphisms are associated with phenotypes related to affective disorders.

We describe 4 complementary studies, assembled to provide replication and to clarify what aspects of circadian polymorphisms may be relevant to both bipolar and unipolar affective disorders.

In four different analyses, circadian gene polymorphisms were studied for association with three phenotypes: bipolar disorder, unipolar depression (major depressive diagnosis or rating-scale quantitative trait), and morningness-eveningness (which shares comorbidity with major depression) 19. In 294 tests of association, 23 different associations met the nominal significance criterion of P < 0.05, whereas 15 such associations would have been anticipated by random chance. It is plausible that most of the nominal associations were due to random chance (false discovery), but at least 5 appeared associated with these affective phenotypes with sufficient evidence of reliability to be considered suggestive.

The intronic SNP rs2314339 in NR1D1 met false discovery and empirical family-wise criteria for significant association with the transmission of bipolar disorder (P = 0.0005). The TDT analysis should be free from biases due to racial stratification. The same SNP was associated with delayed sleep phase disorder cases in an unpublished case-control sample, but significance was not sustained in the case-control sample after preliminary control for racial stratification. In both analyses, the more common allele was associated with the disorder and the less common allele was associated with control or normal health. The DSPD data offered at best only an indirect kind of replication because of the difference in phenotype between bipolar disorder and DSPD, especially considering that in our DSPD sample, we had found comorbidity with unipolar depression but not with bipolar disorder 19. Moreover, rs2314339 has been tested in some whole genome association studies of bipolar disorder, but we are unaware that any suggestive association has been found in such studies. Another nuclear receptor, NR2E1, has been reported to be associated with bipolar disorder in a case-control study 38: there has been brief mention that NR2E1 and NR1D1 may interact in the development of photoreceptors.

NR1D1 (OMIM 602408) is a key element of a unique circadian feedback loop, in which it inhibits transcription of ARNTL (BMAL1) by inhibitory binding at ARNTL RORE promoter sites 39. NR1D1 may similarly inhibit transcription of CLOCK and NPAS2, the proteins of which activate ARNTL by binding as heterodimers. NR1D2 possibly has a similar role (OMIM 602304). These mechanisms may be particularly relevant to bipolar disorder, since it has been suggested that a mutation of CLOCK (which produces hyperactivity) may be an animal model for bipolar mania 40. A knockout of ARNTL, on the other hand, reduces activity, though that can be largely restored by replacing ARNTL function in muscle 41. ARNTL heterodimers with NPAS2 may bind to promoter eboxes of MAOA, thus promoting inactivation of dopamine, and thus inhibiting the pro-manic effects of dopamine 424344. Oddly enough, the ARNTL-CLOCK heterodimer was not demonstrated to have a similar effect on MAOA, though ARNTL-CLOCK would be expected to act on the same e-boxes in the MAOA promoter. Lithium, a primary medication for treatment of bipolar disorder, promotes degradation of NR1D1 through inhibition of GSK3, whereas GSK3 phosphorylation stabilizes NR1D1 45. These interactions are modeled in Fig. 1.

NR1D1 or Rev-erb-alpha is so-called, because it is transcribed in the reverse direction and overlaps the 3' end of THRA, an important thyroid nuclear receptor. It is interesting that thyroid dysfunction becomes most prominent among bipolar patients after treatment with lithium 46. Also, thyroid (T3) augmentation is useful for treating depression 47. Thyroid has been used for periodic catatonia (perhaps a form of bipolar disorder) since the 1930's 1. Considering that the intronic location of rs2314339 indicates no obvious functional role, we suspect that this SNP might be in linkage disequilibrium with some nearby polymorphism with a key functional effect. As linkage disequilibrium for rs2314339 extends through most of NR1D1 and to the 3' end of THRA, the functional element could plausibly be situated in either gene.

A coding SNP in PPARGC1B, Pro203Ala, rs7732671, was over-transmitted to bipolar probands with P < 0.05 and odds ratio 1.55. By itself, we might regard this isolated finding as statistically unimpressive and plausibly a false positive. However, the same SNP was associated with unipolar depression with an odds ratio of 2.12 (P < 0.025). The nominally significant association in both completely separate subject sets with odds ratios in the same direction provides suggestive evidence for a reliable association, especially since neither statistical result is sensitive to false positive results from population stratification. The common allele of this SNP has been associated with obesity (OMIM 608886). A paralogue gene, PPARGC1A, is a regulator of ARNTL and additionally functions through regulation of NR1D1 and NR1D2 effects on ARNTL 48. PPARGC1A possibly binds to RORE sites both on NR1D1 and on ARNTL, and PPARGC1B may act similarly, perhaps providing a partial explanation for effects on both mania and depression, seeming opposites which are both aspects of bipolar disorder (Fig. 1). Thus, there may be a convergence of pathways. A number of SNPs in PPARGC1B achieved nominal significance in a case-control study which included some of these same bipolar subjects, but none approached Bonferroni criteria 22.

The TDT association of 6 CLOCK SNPs with bipolar disorder was intriguing, and appeared consistent with the claim that a CLOCK mutation in mice produces a mouse analog of mania 40. Although nominally significant, associations with these SNPs could all represent false-positive statistical findings. One of the 6 SNPs, rs6850524 was also found to be suggestively associated with bipolar disorder in analyses using a partially-overlapping subject sample 17. Also, CLOCK SNP rs2412648 (P < 0.05 by Chi Square, P > 0.05 by EMP1 in our sample) was part of a suggestively-associated haplotype 17. It would be conceivable that many SNPs in the CLOCK gene, each with a small effect impairing the gene, could in combination make a substantial contribution of bipolar susceptibility. Because the CLOCK gene, extending over roughly 114,338 base pairs, displays high linkage disequilibrium throughout its considerable length, it is possible that the most functional polymorphism has not yet been recognized. Similarly, though various associations with several bipolar phenotypes such as recurrence rates and sleep disturbances have been reported with rs1801260, the T3111C SNP in the 3'UTR region of CLOCK 3637, it is possible that rs1801260 is not the most functional polymorphism in linkage disequilibrium. In our TDT analysis and in BALM studies, rs1801260 was not associated with bipolar disorder, nor was it associated with the morningness-eveningness dimension, as had been reported in other data 36.

In 1978, at a time when the gene causing the Drosophila PER mutant had not yet been identified and the presence of 3 human homologues was unknown, the first author hypothesized that bipolar disorder might be caused by mutation of a homologue of the PER gene 49. The three SNPs in PER2 and rs2585405 in PER1, which were over-transmitted or under-transmitted to bipolar probands with nominal significance, gave weak support to this archaic hypothesis, but certainly suggested no major role for the PER genes in bipolar disorder. On the other hand, associations of affective symptoms with other polymorphisms provided some of the strongest evidence that the circadian system has a role in affective disorders. The lack of consistency in results for the PER homologues in different groups was somewhat disappointing and reminds us that these associations may be false positive results. It would appear that additional large and independent samples must be studied to determine if the PER genes have a real role in human affective disorders. Bright light, which promotes mania, tends to promote transcription of the PER1 and PER2 genes 50, which may then inhibit the action of the ARNTL-NPAS2 heterodimer in stimulating MAOA (Fig. 1). Thus, this pathway may also be consistent with our model in Fig. 1. However, the model ignores numerous problems and complexities of circadian regulation and fails to incorporate the dynamic circadian fluctuations or the photoperiodic interactions among circadian system components. We also have not explained how the observed differences in genetic background might produce the distinctive phenotypes among patients with unipolar and bipolar disorders, nor have we suggested the mechanism by which the same genetic background creates susceptibility in bipolar patients to both mania and depression.

The TEF promoter SNP rs738499 had a statistically unimpressive (P_nominal = 0.023) association with unipolar recurrent depression, with the less common G allele being associated with healthy mood. A similar association with the QIDS-SR in the Sleep Center sample provided a degree of replication (P < 0.001), using independent subjects, independent methods, and a different lab and assay procedure. We had selected rs738499 for genotyping based on a report that rs5996091, a SNP over 500,000 nucleotides remote from TEF, was highly correlated with TEF expression with R² = 0.43 51, and we had noted that of HapMap SNPs nearer TEF, rs738499 had the highest linkage with rs5996091 as well as a likely location in the promoter. By stimulatory binding at D-box promoter sites, TEF may promote transcription of NR1D1, NR1D2, and the PER genes, actions which might be supposed to stimulate mania or counter depression (Fig. 1). However, we do not know if the rs738499 minor allele promotes or inhibits TEF transcription.

Two intronic SNPs in CSNK1E were associated with the BALM among bipolar subjects, meeting Bonferroni criteria for significance. The two SNPs were in virtually perfect linkage disequilibrium with each other. However, among 26 SNPs in CSNK1E identified by resequencing 12,173 nucleotides of exonic, intronic, and promoter regions of the gene, no other SNPs were found to be in substantial linkage disequilibrium with these two SNPs (see Core Nucleotide Report EF015901). Surprisingly, although 4 of the 6 participants from the bipolar sample with these two SNPs displayed extreme morningness, one subject of the 6 had a score at the opposite end of the morningness-eveningness scale. Both SNPs were under-transmitted to bipolars (NS). CSNK1E phosphorylates several of the circadian proteins including the PER proteins and ARNTL, and may even have differential effects on phase adjustment depending on which phosphorylation sites are intact on various substrates 5253. CSNK1D has a somewhat similar role. Note that one CSNK1D SNP was nominally associated with bipolar disorder, and the CSNK1D region on 17q achieved a maximum LOD score of 3.63 in a bipolar association study 54. However, since the two CSNK1E SNPs and the PER3 SNP associated with the BALM scale of morningness-eveningness were not associated with the bipolar and unipolar psychiatric phenotypes, it seems that the circadian polymorphisms which are related to affective disorders do not influence affective state simply through effects on circadian phase, e.g., Fig. 1 does not suggest that the effects are mediated through circadian phase change.

It is important to review some suggestive findings reported elsewhere which were not replicated in these analyses. We were not able to confirm the suggestive evidence we had earlier reported that haplotypes in ARNTL and PER3 were associated with bipolar disorder 15. Those haplotype associations had previously fallen short of Bonferroni criteria. We have also examined certain candidates for association proposed by Mansour and colleagues 1316. These included rs7107287, rs4757142, and rs1982350 in ARNTL, rs2859387 in PER3, and rs2291738 and rs2279665 in TIMELESS. None of these reached nominal significance of P < 0.05 in our TDT analyses. The rs11541353 SNP in NPAS2 and rs2290035 in ARNTL, reported to be associated with seasonal affective disorder 5556, were not significantly associated with bipolar disorder in our families by TDT. In PER3, rs10462020 was not associated with bipolar disorder or unipolar depression. Also, we have not yet demonstrated any association of the PER3 repeat described by Archer et al. with affective disorders 26.

In summary, we found several suggestive associations of circadian gene polymorphisms with affective disorders, some of which we were able to partially replicate. The association of NR1D1 rs2314339 with bipolar disorder and DSPS and the association of PPARGC1B Pro203Ala, rs7732671, with both bipolar and unipolar affective disorders appear the most likely to prove reliable. The association of TEF rs738499 with unipolar depression may also prove reliable. Two intronic SNPs in CSNK1E were associated with the BALM in bipolars, but the inconsistent directions of association stimulate some reserve, and these SNPs were not directly associated with affective diagnoses or symptoms. Each of these leads should be pursued. When genotyping of ancestry-informative markers becomes available, our correlations of the BALM with genotypes should be controlled for population stratification, which is a potential problem in the Sleep Center sample as well. Replication and extension of these results in larger independent samples is needed before the importance of circadian polymorphisms in affective syndromes can be verified. If the findings are confirmed, they will suggest that bipolar and unipolar affective disorders have at least one common genetic susceptibility factor, but several which are distinct.

JRK is a founder and holds equity in Psynomics, Inc. The terms of this arrangement have been reviewed and approved by UCSD in accordance with its conflict of interest policies. The other authors declare that they have no competing interests.

DFK suggested the primary hypotheses, supervised recruitment of the Sleep Clinic sample, selected the polymorphisms to be assayed, and wrote much of the manuscript. CMN contributed to the hypotheses and design of the study, performed most of the statistical analyses, and wrote parts of the manuscript. EJJ assembled the data for the non-related bipolar sample and critiqued the manuscript. TS helped manage and store the DNA samples, performed the SNPlex assays, assembled assay results, and wrote portions of the manuscript. JRK developed the team collecting and assembling the UCSD bipolar samples, participated in the NIMH Bipolar Disorder Genetics Initiative, developed the assay laboratory, contributed to design, and wrote parts of the manuscript. All authors read and approved the final manuscript.