Genetic Basis of Chronotype in Humans: Insights From Three Landmark GWAS

Abstract

Study Objectives:

Chronotype, or diurnal preference, refers to behavioral manifestations of the endogenous circadian system that governs preferred timing of sleep and wake. As variations in circadian timing and system perturbations are linked to disease development, the fundamental biology of chronotype has received attention for its role in the regulation and dysregulation of sleep and related illnesses. Family studies indicate that chronotype is a heritable trait, thus directing attention toward its genetic basis. Although discoveries from molecular studies of candidate genes have shed light onto its genetic architecture, the contribution of genetic variation to chronotype has remained unclear with few related variants identified. In the advent of large-scale genome-wide association studies (GWAS), scientists now have the ability to discover novel common genetic variants associated with complex phenotypes. Three recent large-scale GWASs of chronotype were conducted on subjects of European ancestry from the 23andMe cohort and the UK Biobank. This review discusses the findings of these landmark GWASs in the context of prior research.

Methods:

We systematically reviewed and compared methodological and analytical approaches and results across the three GWASs of chronotype.

Results:

A good deal of consistency was observed across studies with 9 genes identified in 2 of the 3 GWASs. Several genes previously unknown to influence chronotype were identified.

Conclusions:

GWAS is an important tool in identifying common variants associated with the complex chronotype phenotype, the findings of which can supplement and guide molecular science. Future directions in model systems and discovery of rare variants are discussed.

Statement of Significance

An individual’s tendency to be a night owl (late chronotype) or a morning lark (early chronotype) is influenced by his or her genetic makeup. Expression of chronotype (across early and late types) is normally distributed in the population, which suggests that many common genetic variants with modest effects contribute to the phenotype. In our review, we discuss 3 recent hallmark studies using genome-wide association analyses for the identification of genes associated with trait chronotype. Findings identified a number of genes with previously known roles in the circadian clock, as well as several novel gene loci that had not been previously identified as having a role in chronotype or circadian rhythms. Genome-wide analysis is an important tool for identifying novel common gene variants, which illuminates previously unknown genetic influences. Future research on the genetic basis of chronotype should consider using model systems to elucidate circadian function of novel loci.

INTRODUCTION

Circadian rhythms are cyclical changes in cellular, molecular, and biological processes that repeat approximately once every 24 h. Driven by an internal “master clock” located in the suprachiasmatic nucleus (SCN) and influenced by environmental stimuli/zeitgebers (e.g., light–dark cycles), circadian rhythms play a central role in the regulation of many aspects of physiological processes, including sleep–wake cycles. Typical endogenous circadian rhythmicity in humans tends to cycle across a period of 24.2 h with little day-to-day variation in length.¹ The timing of the circadian system, however, varies considerably across individuals. Chronotype, or diurnal preference, refers to behavioral patterns or manifestations indicative of underlying circadian-governed biological processes.² In the study of human sleep–wake regulation, chronotype represents preferred timing of sleep and wake (independent of environmental factors, such as work schedules) and corresponds to the timing of the circadian system.

The timing of circadian rhythmicity falls on a continuum, at the either end of which are individuals colloquially referred to as morning larks and night owls. Morning larks are considered to have early chronotype or advanced sleep phase (also referred to as morning type/preference). Early chronotypes prefer to awaken early in the morning and feel most active during the earlier parts of the day. Some evidence suggests that short circadian periods (e.g., <24 h) may play a role in extreme advanced sleep phase.³ Night owls, or late chronotypes with delayed sleep phase (also evening type/preference), prefer to awaken much later in the day and typically feel most active and motivated in the late evening or at night. In direct contrast to early types, some evidence suggests that individuals with extreme delayed sleep phase may have circadian periods longer than 24.2 h.⁴ Individuals who fall between the ends of the continuum are referred to as intermediate types. Epidemiological studies characterize chronotype as near-normally distributed with robust developmental changes in diurnal preference across the life span, i.e., earlier preference during childhood development, later in adolescence and early adulthood, then progressively earlier as age advances.^2,5

GENETICS OF THE CIRCADIAN CLOCK

There has long been an interest in the genetic basis for circadian rhythmicity. Tremendous advances have been made in identifying a series of molecules that comprise a core circadian clock that exists in every cell, rather than being exclusive to the SCN. Each cell maintains its own circadian rhythmicity, and the master clock in the SCN seems to play the role of maintaining synchronization among these cellular clocks. Primarily based on work in model systems, it is now understood that the molecular circadian clock consists of a negative feedback loop involving the Period (PER1, PER2, and PER3) and Cryptochrome (CRY1 and CRY2) genes.⁶ Other genes involved in the molecular generation of circadian rhythms include: Casein Kinase 1δ and 1ε (CK1) and transcription factors Circadian Locomotor Output Cycles Kaput Protein (CLOCK), Brain and Muscle ARNT-like Protein (BMAL1 and BMAL2), and Neuronal Pas Domain Protein (NPAS1 and NPAS2); for greater in-depth discussion of circadian clock genes, see Lowrey and Takahashi,⁶ McClung,⁷ and Pack et al.⁸

GENETICS OF CIRCADIAN TIMING

In parallel with the identification of the molecular mechanisms of circadian rhythms, studies sought to determine whether chronotype, as a behavioral manifestation of circadian rhythms, is also determined by genetic factors by exploring its heritability. Several twin and family studies have estimated the heritability of chronotype to be approximately 50% based on estimates in the United States,⁹ United Kingdom,¹⁰ Scandinavia,^11,12 and Brazil,¹³ whereas studies of Hutterites¹⁴ and Amazonians¹⁵ suggest lower heritability at 14% and 23%, respectively. Taken together, heritability estimates suggest that genetic factors explain a considerable proportion, up to 50%, of the population variability in circadian timing. Furthermore, the normal distribution of chronotypes indicates a polygenic basis such that multiple genes contribute modest, aggregative effects to this complex phenotype.¹⁶ Given the heritability of circadian typology and prior success in identifying the components of the molecular circadian clock, chronotype is a logical target for genetic studies.

Several studies have used a candidate-gene approach and examined the associations between circadian genes and either chronotype or circadian rhythm sleep disorders. An association was identified and supported between the 3111C allele of the CLOCK gene and eveningness,^17,18 though some studies failed to replicate this finding.^19–21 Associations have been found between extreme delayed sleep phase and length polymorphisms of the PER3 gene^22–24 in addition to the 3111C allele of the CLOCK gene.¹⁷ Morningness has been found to be associated with polymorphisms in the clock genes PER1 and PER2 in other studies.^25,26 In addition, polymorphisms in the PER3 and ARNTL2 genes are associated with chronotype in a sample of 966 British adults.²⁷

Perhaps, the most studied human gene variant with respect to its role in diurnal preference is a PER3 polymorphism with variable number of tandem repeats (VNTR) of a 54-bp motif in exon 18 that is repeated either 4 or 5 times (see Dijk and Archer²⁸ for review of PER3). This polymorphism is found only in primates, for which humans can be homozygotic (4/4 or 5/5) or heterozygotic (4/5) for the number of repeats.^29,30 Investigations of PER3 and chronotype have offered conflicting results such that the 5-repeat allele has been found to be positively associated with morningness and earlier dim-light melatonin onset,^24,31,³² but has also been shown to be elevated in individuals with extreme delayed sleep phase,³³ whereas other studies have shown no association between the PER3 VNTR and diurnal preference^34–37 nor in melatonin or cortisol as objective markers of circadian rhythms.³⁸

Studies of extreme chronotype expression can yield highly informative results. Research on familial advanced sleep phase syndrome (FASPS) has identified specific genetic variants that segregate with the disorder. Linkage analysis, a statistical gene mapping approach used in affected families, has identified a rare missense mutation (i.e., replacement of a single nucleotide) in the clock component gene, PER2, which alters the circadian period and is associated with FASPS.³⁹ However, this finding was not replicated in 2 FASPS pedigrees in Japan,⁴⁰ which highlights the difficulty of replicating results of rare mutations and the limited generalizability of findings concerning extreme phenotype expressions.

Although results from these genetic studies are mixed, overall patterns suggest that multiple genes play important roles in influencing circadian typology. As chronotype is a complex trait, a large number of genes with modest influence contribute to phenotypic manifestation. The detection of multiple small genetic effects requires very large sample sizes, and genome-wide discovery approaches are necessary to identify novel genetic loci. Importantly, going beyond classically gene-centric approaches is essential to elucidate unexplained heritability that remains.

GENOME-WIDE DISCOVERY

The first genome-wide analysis of circadian-related phenotypes analyzed data from 749 subjects in The Sleep Heart Health Study.⁴¹ The authors employed linkage analysis that is used to map genetic loci associated with a phenotype based on observations of related individuals. Linkage analysis for habitual bedtime revealed no significant genetic associations although statistical power was likely limited by insufficient sample size for genome-wide analysis (as large samples are required to detect variants of even moderate effect, i.e., odds ratio [OR] = 1.5–2.0)⁴² and poor quality of genetic data. Notably, near-significant genetic associations for habitual bedtime were observed near CLOCK, PROK2, CSNK2A2, and NSPR1, all known circadian clock-related genes. However, 3 recent large-scale genome-wide association studies (GWASs) of subjects of European ancestry from the 23andMe cohort⁴³ (23andMe, Inc., Mountain View, CA, USA), and the UK Biobank^44,45 have uncovered novel gene loci associated with diurnal chronotypic preference. Here, we will compare methodologies and significant findings of each GWAS (overlap and divergence), closing with a discussion of the findings in the context of prior knowledge in this area.

Unlike candidate gene approaches, GWAS does not require any a priori knowledge of genes underlying the expression of the phenotype of interest. Indeed, GWAS is a discovery approach wherein the entire genome is searched for small genetic variations that occur more commonly in individuals with a particular phenotypic trait. Specifically, a GWAS tests associations between single nucleotide polymorphisms (SNPs; i.e., the most common type of genetic variation among people, represented by a difference in a single DNA base pair) and a phenotype of interest (often a trait or disease). GWASs are typically restricted to common genetic variants (i.e., with minor alleles occurring in more than 5% of the population), which tend to have small or modest effects on complex phenotypes. Because SNPs are often correlated with other nearby SNPs, a substantial proportion of the variability in the genome is captured by measuring several hundred thousand SNPs and imputing millions of non-genotyped SNPs.^46,47

GWAS analyses involve conducting individual association models of the relationships between each SNP and the phenotype of interest (e.g., linear regression models for continuous phenotypes or logistic regression models for binary phenotypes). Due to the large number of tests resulting from all SNPs being examined individually, a multiple-comparisons correction is necessary to determine statistical significance and protect against false-positive associations. While there are different methods for performing this correction, a p-value less than 5 × 10^–8 is typically required to claim genome-wide significance.^47–50

The first large-scale GWAS of diurnal preference was published by Hu and colleagues⁴³ from a European-ancestry sample of worldwide customers of 23andMe, a direct-to-consumer genotyping service. Subjects ranged from early adulthood (<30 years) to old age (>60 years; exact age range not reported). Subjects in this study responded to 2 survey items regarding natural tendency of being a night person/night owl vs. morning person/early bird. Specifically, subjects were asked twice: Are you naturally a night person or a morning person? with a unique set of response options for each administration: (1) night owl, early bird, or neither, then (2) night person, morning person, neither, it depends, or I’m not sure. Responses indicating intermediate type or ambiguity (viz. neither, it depends, and I’m not sure) were treated as missing data. Thus, those who indicated intermediate type on both items were not included in analyses, whereas those who indicated intermediate type on 1 item but also early or late type on another were classified based on their early/late responses. Subjects with diametrically opposed responses, i.e., indicating early chronotype on 1 item and late chronotype on another, were excluded from analyses entirely, resulting in 89283 unrelated individuals (38937 morning types, 50346 evening types) included in the analysis. The primary chronotype phenotype in this study was a binary outcome value (early vs. late chronotype), and analyses included as covariates age and sex. In addition, the authors adjusted for the top 5 principle components accounting for population stratification that controls for differences in allele frequencies among populations of different ancestry. The authors identified 15 loci associated with self-rated diurnal preference (Table 1, see Figure 1 for Manhattan plot that shows the genome-wide significant SNPs and nearby loci). Although Hu et al. conducted a follow-up GWAS with chronotype scored on a continuum including intermediate types, SNP association results from this analysis were only provided for 2 newly significant SNPs (designated by asterisks [*] in Table 1). Notably, because chronotype is a continuous trait, phenotyping chronotype on a continuum yields greater statistical power to detect genetic associations than operationally defining chronotype as a binary trait.

Table 1.

Gene Loci Associated With Chronotype/Diurnal Preference in 23andme And UK Biobank Genome-wide Association Analyses.

SNP	EAF	Chr:Pos	Nearest gene(s)	β	Cont P	OR	Bin P	Study	Known CRS function
Loci identified in all 3 GWASs
rs12736689	0.03	1:182,549,729	RGS16, RNASEL	—	—	0.74	7.00E-18	Hu	K
rs1144566	0.97	1:182,569,626	RGS16	−0.10	2.62E-14	0.74	1.29E-08	Lane	K
rs516134	0.03	1:182,553,693	RGS16	0.08	9.00E-13	1.21	3.00E-12	Jones	K
rs55694368	0.93	2:239,317,692	PER2	—	—	0.86	2.60E-09	Hu	K
rs35333999	0.043	2:239,161,957	PER2	0.06	8.43E-08	1.21	9.01E-06	Lane	K
rs75804782	0.88	2:239,316,043	PER2	0.03	3.00E-07	1.09	4.00E-10	Jones	K
rs3972456	0.29	7:102,436,907	FBXL13, FAM185A		—	0.92	6.00E-09	Hu	P
rs372229746	0.45	7:102,158,815	FBXL13	0.03	5.18E-10	1.12	4.29E-07	Lane	P
rs372229746	0.55	7:102,158,815	FBXL13, ORAI2, RASA4	0.03	4.00E-09	1.06	7.00E-07	Jones	P
rs10493596	0.24	1:77,713,434	AK5		—	1.09	8.00E-12	Hu	U
rs76681500	0.84	1:77,726,241	AK5	–0.04	1.50E-12	0.86	1.77E-09	Lane	U
rs11162296	0.16	1:77,700,196	AK5, PIGK	–0.04	2.00E-12	0.93	1.00E-12	Jones	U
Loci identified in 2 GWASs
rs35833281	0.21	6:55,021,561	HCRTR2	—	—	0.92	2.60E-09	Hu	K
rs76899638	0.22	6:55,147,508	HCRTR2	0.03	4.00E-08	1.05	2.00E-07	Jones	K
rs2050122	0.8	1:19,989,205	HTR6	0.03	4.61E-08	1.12	7.94E-07	Lane	K
rs2050122	0.2	1:19,989,205	HTR6	0.03	2.00E-08	1.06	3.00E-06	Jones	K
rs4821940	0.55	22:40,659,573	TNRC6B	0.03	1.05E-08	1.07	8.56E-05	Lane	P
rs4821940	0.45	22:40,659,573	TNRC6B, SGSM3	0.02	3.00E-08	1.05	4.00E-08	Jones	P
rs34714364	0.17	1:150,234,657	APH1A, CA14	—	—	1.12	2.00E-10	Hu	P
rs10157197	0.4	1:150,250,636	APH1A	0.03	1.48E-09	1.13	1.27E-11	Lane	P
rs12635074	0.65	3:55,982,416	ERC2	–0.03	3.08E-08	0.92	1.20E-06	Lane	U
rs11708779	0.32	3:55,934,939	ERC2	–0.02	3.00E-08	0.96	2.00E-06	Jones	U
Loci identified in 1 GWAS
rs1015197	0.6	1:150,250,636	PRPF3, TARS2	0.03	1.00E-09	1.05	5.00E-07	Jones	U
rs11121022	0.42	1:7,836,659	PER3, VAMP3	—	—	1.07	2.00E-08	Hu	K
rs72720396	0.23	1:91,191,582	CALB1	–0.03	1.00E-07	0.95	3.00E-08	Jones	P
rs12140153	0.9	1:62,579,891	INADL	0.04	7.00E-09	1.07	4.00E-06	Jones	P
rs141175086	0.998	1:780,397	LINC01128	–0.27	1.42E-04	0.46	4.38E-08	Lane	U
rs1075265	0.52	2:54,354,927	PSME4, ACYP2	–0.03	2.00E-10	0.95	4.00E-08	Jones	U
rs70944707	0.23	2:24,257,444	FKBP1B	0.03	3.00E-08	1.05	2.00E-05	Jones	U
rs11895698	0.14	2:239,338,495	ASB1	0.04	1.15E-08	1.1	1.30E-04	Lane	U
rs1595824	0.49	2:198,874,006	PLCL1	—	—	1.08	1.20E-10	Hu	P
rs148750727	0.995	4:188,022,952	FAT1	0.15	3.61E-06	2.34	1.58E-08	Lane	U
rs9479402	0.01	6:153,135,339	VIP	—	—	0.69	3.90E-11	Hu	K
rs9357620*	0.71	6:13,170,634	PHACTR1	0.02	1.70E-08		—	Hu	U
rs2948276	0.88	7:96,457,119	DLX5, SHFM1	—	—	0.92	1.10E-08	Hu	U
rs2975734*	0.43	8:10,030,097	MSRA	0.02	2.10E-09		—	Hu	K
rs192534763	0.99	8:36,202,946	UNC5D	0.1	3.00E-07	1.25	2.00E-08	Jones	U
rs17311976	0.19	8:131,637,337	ADCY8	0.19	1.08E-06	1.13	3.37E-08	Lane	P
rs77641763	0.88	9:140,265,782	EXD3, GRIN1, NRARP	0.04	5.00E-11	1.07	7.00E-09	Jones	U
rs6582618	0.52	12:38,726,137	ALG10B	—	—	1.07	1.50E-08	Hu	U
rs542675489	0.6	12:120,994,888	RNF10	0.03	3.29E-09	1.09	1.38E-05	Lane	U

SNP = single nucleotide polymorphism. EAF = effect allele frequency. Chr = Chromosome. Pos = position. Cont P = significance for genetic associations in GWAS on chronotype as a continuous phenotype. β = beta coefficient, i.e., effect size for genetic associations in GWAS on chronotype as a continuous phenotype. Bin P = significance for genetic associations in GWAS on chronotype as a binary (cases vs. controls) phenotype. OR = odds ratio, i.e., effect size for genetic associations in GWAS on chronotype as a binary phenotype. Known CRS function = whether gene has a previously known association with circadian rhythms or sleep. K = Known. P = Probable. U = Unknown.

Figure 1.

Manhattan plot of genome-wide association studies of being a morning person from Hu et al. 2016.⁴⁴ Presented in its originally published format, the grey line corresponds to threshold for genome-wide significance at p = 5 × 10^–8, with significant results shown above this threshold in red. Gene labels are annotated as the nearby genes to the significant single nucleotide polymorphisms.

Lane et al.⁴⁴ and Jones et al.⁴⁵ each conducted genome-wide association analyses for chronotype on data from British, Caucasian subjects from the UK Biobank, a prospective study of >500000 people in the United Kingdom. Individuals in the UK Biobank range in age from 37 to 73 years although Lane et al.⁴⁴ analyzed data only from subjects between the ages of 40 and 69 years to increase sample homogeneity. In these studies, GWASs were conducted for chronotype defined as both a binary and a continuous trait, which was self-reported in response to the single item from the Morningness–Eveningness Questionnaire (MEQ) that explains the highest fraction of variance in preferences of sleep–wake timing:^51,52Do you consider yourself to be… responses to which included self-description as definitely a morning person, more a morning than evening person, more an evening person than a morning person, definitely an evening person, do not know, and prefer not to answer (subjects endorsing prefer not to answer were excluded from analyses in both studies).

From their GWAS analyses, Lane et al.⁴⁴ excluded shift workers, subjects taking sleep medications, and those who responded do not know to the chronotype item, resulting in a sample size of n = 100420. The authors conducted 2 parallel GWASs: a continuous chronotype outcome (1–4, definitely a morning person to definitely an evening person) and a binary comparison of the 2 extreme phenotypes (26948 definite morning types vs. 8724 definite evening types). Genetic association analysis was performed using a SNPTEST while controlling for age, sex, and 10 principal components of ancestry and genotyping array (2 different, though highly similar, Affymetrix arrays were used for different subjects in the UK Biobank). The authors identified 12 significant loci and 1 suggestive secondary signal associated with chronotype (Table 1).

Jones and colleagues⁴⁵ also conducted parallel GWASs for continuous and binary operationalizations of chronotype. However, their continuous chronotype was coded differently such that the response of do not know served as a proxy for intermediate type, rather than excluded as missing data. Thus, their continuous variable included 5 groups (i.e., definite morning, moderate morning, intermediate, moderate evening, definite evening types) rather than 4 as in the Lane et al study. Chronotypes were then normalized for analysis through adjustment of the raw phenotype for age, gender, and study center, followed by inverse normalization of the resulting residuals. Unlike Lane et al who compared just extreme phenotypes in their binary GWAS, Jones et al collapsed across definite and moderate groups within morning and evening types, resulting in a sample size of 114765 in the GWAS of binary chronotype (compared to n = 35672 in the GWAS of extreme phenotypes in Lane et al.). In addition, exclusionary criteria differed from the prior 2 investigations such that Jones et al. excluded patients with diabetes who reported insulin use within the first year of diagnosis, patients with diabetes diagnosed younger than the age of 35 years or with no known age of diagnosis, and patients with diabetes diagnosed within a year of the study. These exclusions were due to the authors' focus on the relationship between chronotype and diabetes. The authors used linear mixed models to conduct GWAS analyses to account for population structure and subject relatedness, while controlling for age, sex, study center, and genotyping array. Jones et al.⁴⁵ identified 16 significant loci associated with chronotype in the UK Biobank (Table 1).

All 3 GWASs independently supported associations with chronotype for genes PER2, RGS16, FBXL13, and AK5 based on nearby significant SNPs. Of these identified genes, PER2 and RGS16 are both known for their roles in circadian regulation. The degradation of PER2, the speed of which influences circadian period length, is governed by Casein kinase 1’s phosphorylation of DEC2 (BHLHE41).⁵³ A regulator of G-protein signaling 16, RGS16, has been shown to regulate G_αi/o-mediated cyclic adenosine monophosphate synthesis in the SCN. RGS16 deficiency/ablation leads to delayed dorsomedial PER1 expression resulting in circadian period lengthening in mice, highlighting the association between chronotype and RGS16.⁵⁴ The F-Box protein, FBXL13, has a suspected circadian influence,^55–57 whereas AK5 had not previously been linked to chronotype or circadian processes.⁵⁸ Regarding FBXL13, the SCF^FBXL3 complex regulates expression of CRY and PER through ubiquitination and degradation of CRY proteins via the SCF^FBXl3 ubiquitin ligase complex, and mutations in FBXL3 are associated with lengthened circadian periods in mice.^55–57 Adenylate kinase 5, i.e., AK5, has been shown to regulate the adenine nucleotides within cells⁵⁸ but has not been tied to chronotype or the circadian pacemaker prior to these studies.

Five additional genes were found to be associated with chronotype in 2 of the 3 studies (Table 1). Hypocretin (orexin) Receptor Type 2 (HCRTR2) is a gene that regulates the expression of the G protein–coupled hypocretin receptor type 2. Deficient hypocretin systems are thought to be associated with a reduction of wakefulness and circadian alertness signals in narcolepsy.⁵⁹ 5-Hydroxytryptamine (serotonin) Receptor 6 (HTR6) has not been previously linked to chronotype but is known for its connection to sleep regulation in rats such that 5-HTR₆ agonists lead to reductions in sleep.^60,61 Trinucleotide Repeat Containing 6B (TNRC6B) is a protein-coding gene that belongs to the GW182 family of proteins, which are necessary for micro-RNA gene silencing in animal cells. These proteins have been linked to circadian behavior in Drosophila⁶² and bound to circadian transcription factors in mouse liver,⁶³ although these GWASs offer the first evidence linking this gene to human circadian typology. Both Hu et al. and Lane et al. identified the association of APH1A and chronotype. APH1A encodes a subunit of the γ-secretase complex that cleaves the amyloid-β precursor protein.⁶⁴ Hu et al. highlighted that amyloid-β levels are higher in wakefulness than sleep and are affected by manipulations of the hypocretin (orexin) system,^65,66 in addition to clearance by the glymphatic system of the brain during sleep.⁶⁷ The last gene to be implicated by 2 independent GWASs was ERC2 (ELKS/RAB6-Interacting/CAST Family Member 2), a protein-coding gene belonging to the Rab3-interacting molecule-binding protein family involved in the regulation of neurotransmitter release through involvement in the organization of the cytomatrix at the nerve terminals active zone. This gene has no previously known association with sleep or circadian rhythmicity.

Highlighting the genetic complexity of chronotype as well as the limitations of genome-wide association analysis, 24 SNPs across these 3 GWASs were genome-wide significant in just 1 of the 3 studies (see Table 1 for full list). Unlike the replicated genes discussed above, relationships of these unreplicated genes with chronotype were largely unclear, e.g., ALG10B, LINC01128, and UNC5D. Even so, 5 genes with well-documented circadian and/or sleep regulatory roles—PER3, Vasoactive Intestinal Peptide (VIP),⁶⁸FBXL3, Methionine Sulfoxide Reductase A (MSRA), and RAS Dexamethasone-Induced 1 (RASD1)—were significant in 1 of the 3 GWASs. Given the lack of replication across studies, these associations, particularly for those with no known circadian roles, should be interpreted with great caution.

Importantly, of the 9 genes identified in multiple studies (4 found in all 3 studies and 5 found in 2 studies), only 3 SNPs were shared: rs372229746 in FBXL13, rs2050122 in HTR6, and rs4821940 in TNRC6B. Effects of these shared SNPs on chronotype were in the same direction and exhibited similar magnitudes. Only the studies by Lane and Jones shared SNPs in common, given the usage of the same genotype data from the UK Biobank. Both Lane et al. and Jones et al. attempted to replicate the findings from the 23andMe SNPs reported in the Hu study. In Lane et al.’s GWAS of extreme chronotype, 8 of the 15 significant SNPs identified in Hu et al. were found to have genome-wide significance, and all 15 signals had the same direction of effect in both studies. In a follow-up meta-analysis of both samples, Lane et al. identified 3 additional SNPs near PER3, VIP, and TOX3. In the Jones study, the authors conducted a meta-analysis on their UK Biobank sample and the 23andMe cohort from the Hu study. Findings supported 13 of Jones et al.’s initially identified 16 loci to be associated with chronotype although 2 signals were in the opposite direction.

DISCUSSION

The results of these 3 GWAS reports highlight the potential of discovery-based approaches for the identification of novel candidate genes involved in chronotype, circadian rhythms, and circadian rhythm sleep disorders, while also pointing to the need for large sample sizes to find common genetic variants with small effects for complex phenotypes. In particular, GWAS have demonstrated promise in revealing unexpected genetic effects that have gone undetected in more traditional candidate gene studies.^69,70 Despite some differences in phenotyping and analytic approach, there was a good deal of consistency across studies with 9 genes identified in at least 2 of the 3 main studies reviewed. However, given that 2 studies both used data from the UK Biobank, one might expect the results of these analyses to be more similar. There are several important lessons that can be drawn from these studies.

Importance of Rigorous and Consistent Phenotyping

Multiple methods exist to capture the timing of the circadian system, including assessments of core body temperature, dim light melatonin onset, self-reported diurnal preference, and actigraphy- or diary-based midpoint of sleep.⁷¹ For large-sample research endeavors, non-invasive, cost-effective, and expeditious assessments are needed to capture the phenotypes of the chronotype. At its most rudimentary level, chronotype can be assessed with a single question such as those used by 23andMe and the UK Biobank.¹ While these questions have face validity, sole reliance on a single, subjective item to define the chronotype phenotype may lead to difficulties regarding individual differences in beliefs about what constitutes morning or evening types and fails to capture much of the variability in chronotype within the population.² Measurement error from non-rigorous phenotyping of complex traits decreases the signal-to-noise ratio and attenuates that measure’s ability to demonstrate association with genetic variants, thus reducing power to detect significant genome-wide associations. Furthermore, although circadian rhythms play important roles in its expression, chronotype is influenced by several other genetic and environmental factors, which can add further noise to the phenotype and reduce statistical power. Thus, non-rigorous phenotyping can yield unreliable significant findings that may be less likely to be replicated in subsequent studies.⁷²

In the 3 large GWASs reviewed above, chronotype was determined by 1 or 2 self-report questions. Specifically, Lane et al.⁴⁴ and Jones et al.⁴⁵ used a single item of the MEQ, whereas Hu et al.⁴⁴ measured chronotype using 2 identically worded items (Are you naturally a night person or a morning person?) with 2 similar sets of response options. Although the differences between these items are slight, they likely contributed to disparate findings between results from the 23andMe and UK Biobank samples, highlighting the impact that small variations in phenotype definition can have on genome-wide associations. Moreover, even when the measure for phenotyping was consistent across analyses (both within studies conducting parallel GWASs and across the 2 separate studies analyzing UK Biobank data), differences in response coding (e.g., binary vs. continuous, treatment of ambiguous answers) affect measurement error and yield disparate findings. These are the same issues that plague other epidemiological research and produce varying rates of disease prevalence and incidence. Future endeavors should prioritize more rigorous and consistent phenotyping strategies including the use of standardized self-report instruments, clinical interviews, or objective measures of sleep–wake periodicity (e.g. actigraphy), as GWASs will have greater statistical power and produce more reliable results when phenotypic information is grounded on a solid base of defining criteria. There is also a critical need to validate commonly used self-reported items by comparing them to biological markers of circadian rhythms to determine which question(s) most closely relate(s) to endogenous processes.

Analytic Approaches Yield Differing GWAS Results

Differences in the analysis plan, including decisions defining the sample and selections regarding covariates, statistical tests, and methods to address confounders, may also produce disparate findings across GWASs. The first critical analytic decisions regard inclusion and exclusion criteria. Notably, all 3 GWASs analyzed data from subjects with European Ancestry to increase sample homogeneity and accounted for residual population structure in their models. Where Hu et al.⁴³ and Lane et al.⁴⁴ excluded individuals based on genetic relatedness, Jones et al.⁴⁵ statistically adjusted for relatedness in their analyses. In addition, Lane et al. excluded shift workers and subjects taking sleep medications, as shift work and hypnotics may potentially influence self-reported chronotype. The Jones study, however, excluded individuals based on diabetic and autoimmune considerations (see above).

Where Hu et al. and Lane et al. used regression for their genome-wide analyses, Jones et al employed linear mixed modeling that allowed adjusting for between-subjects genetic relatedness. Furthermore, the Lane study broke from convention by using a less robust “expected” genotype count (also known as the posterior mean⁷³ or allele dosage⁷⁴) in SNPTEST, rather than more traditionally employed frequentist test statistics (e.g., chi-square statistics for case–control studies, the score test for quantitative trait analyses). Lane et al.’s choice of model reflects efforts to account for uncertainty in imputed SNP quality; in particular, difficult-to-impute SNPs, i.e., those with low information scores following imputation. While both models asymptotically approximate each other, frequentist test statistics (using likelihood estimation) tend to result in larger effect sizes for difficult-to-impute SNPs.⁷³ Thus, frequentist test statistics may overestimate genetic effects, which may have influenced why the Lane study chose a different, more conservative, statistical model. Although all studies adjusted for age, sex, and population structure, differences of methods of statistical adjustment and in covariate selection existed across studies. It is unclear whether different estimation techniques (e.g., regression vs. linear mixed modeling) and covariate selection contributed to the disparate results across studies. However, the use of multiple approaches can be beneficial by allowing for sensitivity analysis such that genetic associations that are significant across different strategies are more likely to be true findings.

The Complementary Nature of GWAS and Molecular Studies

Findings from the GWAS analyses confirmed the roles of known circadian genes for influencing chronotype, a behavioral phenotype. Even so, perhaps the most exciting findings from these GWASs center on the significant SNPs near genes with little to no known history of circadian influence, such as AK5, APH1A, and ERC2 among many others. Indeed, although many classical clock genes were significantly associated with chronotype in these GWASs, a critical strength of these exploratory techniques is the ability to identify novel genetic etiology of these complex phenotypes. Owing to discovery of the novel loci and newly implicated genes, efforts should be directed toward replicating these findings and elucidating their roles in circadian regulation and chronotype. Comparison of phenotype-associated variants to expression quantitative trait loci variants can help prioritize a gene and mechanism for functional follow-up.^75,76 As genome-wide discovery attempts to establish the genetic architecture and functional genomic model of circadian rhythms and chronotype, studies using model systems can reveal the molecular mechanisms modulating circadian processes. Molecular studies can be used to support associations identified through genome-wide analysis by examining these identified regions in finer detail and finding a potential mechanism by which they act on chronotype. Indeed, genetic model systems for studying sleep and sleep–wake disorders are needed to reveal the molecular basis of chronotype and circadian typology. Genes with no previously known circadian role should now be investigated in model systems to determine whether these genes do have a hitherto unknown role in circadian regulation.⁷⁷ Importantly, failure to replicate GWAS findings in molecular investigations may reveal erroneous associations from GWAS results.

LIMITATIONS AND FUTURE DIRECTIONS

Common vs. Rare Variants

While genome-wide discovery has greatly advanced our knowledge of the genetic basis of sleep and circadian regulation, genetic variants obtained through genotyping chips used in GWAS are restricted to common SNPs with minor allele frequencies (MAF) greater than 5%. Importantly, common variants identified in GWAS typically result in modest influences on the manifestation of complex phenotypes (Figure 2). Despite very large samples in the 3 GWASs we reviewed, differences in findings across studies are likely influenced by effect allele frequency (EAF) and the effect size that each allele contributes. The graphs in Figure 3 show that the more common an effect allele is in the population (i.e., higher EAF), the greater the phenotypic effect must be to detect a genetic association. This demonstrates that achieving sufficient statistical power to detect effects of common alleles requires a greater phenotypic effect at that allele. The complicating matter, however, is that the frequency of effect alleles and their magnitudes of effect are typically inversely related, that is, the more common an effect allele is, the more modest its effect is likely to be (Figure 2).^78,79 Over the range of effect sizes (OR = 1.04–2.34) noted in GWASs of Hu, Lane, and Jones, it is not surprising that there was divergence between significant alleles, given the sample sizes and group proportions in each of the studies. The next-step research on common alleles contributing to chronotype may utilize polygenic risk scores, which aggregate the multiple small effects of common variants to effectively explain much of the heritability of complex/polygenic traits,⁸⁰ such as prior work on schizophrenia⁸¹ and height.⁸² Derivation of polygenic risk scores would allow for the examination of whether genetic load for late chronotype is associated with phenotypic variations in mood, sleep, or other circadian-linked health outcomes.⁸³

Figure 2.

Relationship between allele frequency and effect on phenotype expression. Common variants with minor allele frequencies (MAF)>5% typically boast weak or modest effects on expression of complex phenotypes. The effects of common variants are often detected using genome-wide association studies (GWASs). In contrast, rarer variants typically wield stronger effects and contribute to more extreme expressions of complex phenotypes. Rare variants, however, are not often detected using genome-wide analysis.

Figure 3.

Comparison of statistical power in the 3 genome-wide association studies (GWASs). Comparisons of the 3 different study designs (number of cases:number of controls) and their power to detect alleles with varying genetic relative risk over the range of effect allele frequencies (EAF). The differences in the power over the range of genotype relative risks clearly indicate the impact of sample size (Jones et al. > Hu et al. > Lane et al.), which figures most prominently at higher allele frequencies. Power estimates were performed using the CaTS power calculator (http://csg.sph.umich.edu//abecasis/cats/)⁸⁹ for all studies, using a significance threshold of 5 × 10⁻⁸, prevalence of morningness phenotype of 0.652 (estimated from total UK Biobank cohort), and EAF minimum (0.01) and minimum (0.998) values used to reflect the range across studies.

In contrast, rare gene variants (i.e., SNPs with minor allele frequencies less than 5%) often exert larger effects on phenotype expression but are not typically identified using genome-wide discovery techniques due to their low frequency (see Pack et al.⁸ for review). Given that chronotype is relatively normally distributed in the population, a large number of common variants with small effects contribute to its expression. Even so, rare variants, quite possibly in the same genes as known common variants, likely result in greater magnitude of influence on chronotype when present. The influence of rare variants is particularly germane to extreme expressions of chronotype and circadian pathology, such as in FASPS.^3,41 Given that rare variants may disproportionately influence a trait at the same loci as the common variants identified, fine mapping of these regions may help elucidate the roles that genes in the region play in chronotype expression, beyond simple linkage disequilibrium (i.e., associations between nearby genetic loci) with plausible sleep-related genes.⁸⁴

Demonstrating the significant impact of rare variants, He et al.⁸⁵ studied 2 genetically related subjects with markedly advanced circadian phase and short sleep compared to other members in their family and the general population. Despite habitual sleep periods of approximately 22:00 to 04:00, these individuals denied sleep-related daytime impairment. After sequencing all clock and clock-related genes, He et al. identified a rare point mutation in DEC2 as complicit in their unique and extreme sleep phenotypes. The authors then replicated the effect of this mutation in transgenic mice and Drosophila. Pellegrino and colleagues extended this line of research by demonstrating that genes with rare variants with large effects may harbor other such mutations. Specifically, Pellegrino et al.⁸⁵ sequenced DEC2 in 2 human samples—one from a twin study⁸⁷ and another from a study on sleep deprivation⁸⁸—and found 2 new mutations in DEC2 associated with short sleep. Taken together, these findings highlight the importance of advancements in genetic sequencing and targeted efforts to identify and characterize rare variants associated with sleep and circadian phenotypes.

Further elucidation of both the common and rare variants constituting the genetic basis of chronotype will provide critical insights into the molecular underpinnings of the circadian system. This knowledge will enhance the understanding of the etiology of circadian rhythm and other illnesses with known circadian abnormalities, including major depression and cardiometabolic disorders. These discoveries may lead to potential novel pharmacologic and non-pharmacologic treatment options that target the circadian system. For instance, the creation of personalized genetic risk profiles for circadian abnormalities may be used to guide the design and assignment of shift work schedules to reduce risk for shift work disorder in the future.

FUNDING

This work was supported by NIH grant T32 HL110952-04.

DISCLOSURE STATEMENT

Dr Gehrman has had research grant support from Merck, Inc. and has served as a consultant for Johnson & Johnson and General Sleep Corporation. The other authors reported no financial conflicts of interest.