Genome-wide association analysis of excessive daytime sleepiness identifies 42 loci that suggest phenotypic subgroups

Excessive daytime sleepiness (EDS) affects 10-20% of the population and is associated with substantial functional deficits. We identified 42 loci for self-reported EDS in GWAS of 452,071 individuals from the UK Biobank, with enrichment for genes expressed in brain tissues and in neuronal transmission pathways. We confirmed the aggregate effect of a genetic risk score of 42 SNPs on EDS in independent Scandinavian cohorts and on other sleep disorders (restless leg syndrome, insomnia) and sleep traits (duration, chronotype, accelerometer-derived sleep efficiency and daytime naps or inactivity). Strong genetic correlations were also seen with obesity, coronary heart disease, psychiatric diseases, cognitive traits and reproductive ageing. EDS variants clustered into two predominant composite phenotypes - sleep propensity and sleep fragmentation - with the former showing stronger evidence for enriched expression in central nervous system tissues, suggesting two unique mechanistic pathways. Mendelian randomization analysis indicated that higher BMI is causally associated with EDS risk, but EDS does not appear to causally influence BMI.

EDS is a chief symptom of chronic insufficient sleep 1 as well as of several primary sleep disorders, such as sleep apnea, narcolepsy, and circadian rhythm disorders 2-5 . Several disease processes and medications also associate with prevalent and incident EDS [6][7][8][9] . EDS is estimated to contribute to risk for motor vehicle crashes, work-related accidents and loss of productivity, highlighting its public health importance 10,11 . The clinical impact of EDS extends to a negative impact on cognition, behavior, and quality of life 12,13 . Therefore, sleep interventions often identify reduction in EDS as a chief goal 14,15 . EDS is also associated with an increased risk for cardio-metabolic disorders, psychiatric problems and mortality 9,16,17 through pathways that may be causal, bi-directional, or reflect pleiotropic effects.
While EDS occurs in a variety of settings associated with insufficient sleep, there is large interindividual variability in levels of EDS that is not fully explained by sleep duration, sleep quality or chronic disease. Experimental studies have shown that there is also individual vulnerability to EDS following sleep restriction 18,19 . The heritability of EDS is estimated to be between 0.37 to 0.48 in twin studies, 0.17 in family studies, and between 0.084 to 0.17 in GWAS 20,21 , suggesting that genetic factors contribute to variation in sleepiness. Despite multiple candidate gene studies [22][23][24]25,26 , including one from the first genetic release of the UK Biobank 21 , few significant genetic variants have been reported, likely reflecting the heterogeneous and multifactorial etiology of the phenotype and low statistical power. Here, we extend our EDS GWAS to the full UK Biobank dataset 27 , enabling identification of multiple genetic variants and molecular pathways that may contribute to EDS.
In the UK Biobank, 27 452,071 participants of European genetic ancestry self-reported frequency of EDS using the question: "How likely are you to doze off or fall asleep during the daytime when you don't mean to? (e.g.: when working, reading or driving)", with the answer categories "never" (N=347,285), "sometimes" (N=92,794), "often" (N=11,963), or "all of the time" (N=29).
The severity of EDS increased with older age, female sex, higher BMI, various behavioral, social and environmental factors, and chronic diseases (Supplementary Table 1). EDS was positively correlated with self-reported insomnia symptoms, morning chronotype, ICD10 or physician diagnosed sleep apnea and self-reported shorter and longer sleep duration, consistent with earlier reports or known clinical correlates 21 (Supplementary Table 1a and 2). EDS was also correlated with shorter sleep duration, lower sleep efficiency (indicating more time awake during the sleep period) and longer daytime inactivity duration estimated using a 7-day accelerometry in a subset (N=85,388) of UK Biobank participants (Methods; Supplementary Table 1b) 20 .

GWAS and replication
We performed a GWAS of EDS treating the four categories as a continuous variable using a linear mixed regression model 28 adjusted for age, sex, genotyping array, ten principal components (PCs) of ancestry and genetic relatedness matrix, and identified 37 genome-wide significant loci (p<5×10 -8 ) (Fig. 1, Supplementary Fig.1

and Supplementary Table 3). Previously
identified loci for EDS in the first release of the UK Biobank (N=111,975) 21 , including loci at or near PATJ, HCRTR2, and CPEB1, were confirmed in this study. The most significant association was observed at KSR2, a gene regulating multiple signaling pathways, including the ERK/MEK signaling pathway, affecting energy balance, cellular fatty acid and glucose oxidation that is implicated in obesity, insulin resistance a n d heart rate during sleep in previous studies in humans and mice 29,30 . Additional novel loci were identified within or near genes with known actions on sleep-wake control regulation or that are associated with sleep disorders (e.g. PLCL1 31 , GABRA2 32 ,BTBD9 33 ,HTR7 34 ,RAI1 35 ), metabolic traits (e.g. GCKR 36 , SLC39A8 37 ) and psychiatric traits (e.g. AGAP1 38 ,CACNA1C 39 ). No association was seen with SNPs highlighted in smaller independent GWAS of EDS, hypersomnia or narcolepsy 20,25,26,[40][41][42][43][44][45] (Supplementary Table 4).
Previous longitudinal research indicated obesity and weight gain were associated with EDS incidence 8 ; therefore, we performed an additional GWAS adjusting for BMI to identify loci that may operate in obesity-independent pathways. This analysis identified 5 additional loci (Supplementary Fig. 2 and Supplementary Table 3). Effect estimates at the 37 loci identified in the primary model were largely unchanged.
Sensitivity analyses adjusting for potential confounders (including depression, socio-economic status, alcohol intake frequency, smoking status, caffeine intake, employment status, marital status, and psychiatric problems) did not substantially alter effect estimates of the 42 identified signals (Supplementary Table 5). Analyses stratified by obesity and sleep duration revealed consistent effect directions (heterogeneity P>0.05) but lower statistical evidence of association, likely reflecting the smaller subgroup sample sizes (Supplementary Table 6 Table 9). Only eight individual signals, including KSR2, were marginally significant (p<0.05) in individual cohorts and/or meta-analysis of these, likely due to the lower power, inconsistent questionnaires across different cohorts and the multi-factorial etiology of EDS (Supplementary Table 10). However, a genetic risk score (GRS) of all 42 EDS loci weighted by the effect estimates from our primary EDS GWAS was associated with increased EDS in a meta-analysis of HUNT, FINRISK and Health 2000 (Fisher's p=0.001; Supplementary Table 10).

Heterogeneous effects of EDS loci on other sleep traits
The 42 EDS-associated genetic variants likely influence EDS through different mechanisms.
Therefore, to dissect heterogeneity, we tested for association of the combined GRS and individual SNPs with subjective and objective measures of sleep patterns and disorders (Table 1 and Supplementary Table 11).
A GRS of all 42 variants was associated with self-reported shorter sleep duration, morning chronotype, increased insomnia risk, increased frequency of daytime napping, and with accelerometry-assessed lower sleep efficiency and increased duration of daytime inactivity (Table 1). The EDS GRS was not associated with 7-day accelerometry-derived continuous sleep duration, which may reflect differences in self-reported habitual sleep vs. accelerometryestimated sleep, different time points in data collection (accelerometry data were collected 5-7 years after questionnaires), lower power given smaller sample size in accelerometry data, or recall or other biases in the self-report such as from reporting in 1 hour increments.
Individual EDS increasing alleles at PATJ and PLCL1 were also associated with morning chronotype (as previously reported 31,49 ); at metabolism regulatory genes KSR2, LOC644191/CRHR1 and SLC39A8 with self-reported sleep duration (KSR2 with increased sleep duration, LOC644191/CRHR1 with long sleep and SLC39A8 with short sleep); LMOD1 and LOC644456/LOC730134 with both insomnia and short sleep duration; and at the orexin/hypocretin receptor HCRTR2 (known to play a central role in sleep wake control and narcolepsy 50 ) with both morning chronotype and short sleep duration, suggesting common genetic factors. Consistently, adjusting for sleep disturbance traits (ICD10 code defined sleep apnea or narcolepsy, or self-reported sleep duration hours, frequent insomnia symptoms or chronotype) together attenuated effect estimates for several loci, suggesting that these genetic variants influence EDS through altered sleep patterns; however, adjustment for any trait alone only minimally altered effect estimates at select individual loci (Supplementary Table 5).
Using 7-day accelerometry-derived data available in a subset of the UK Biobank (N=85,388), we observed associations of several EDS alleles with reduced sleep efficiency (e.g., SNX17) whereas others were associated with increased sleep efficiency (e.g., PLCL1), suggesting that genetic mechanisms may lead to EDS through effects on increased sleep fragmentation (i.e., low sleep efficiency) or increased sleep propensity (i.e., high sleep efficiency), respectively. Therefore, we performed clustering analysis on EDS risk alleles at 42 loci according to their association effect sizes (z-scores) with objective estimates of sleep efficiency, sleep duration, and number of sleep bouts, and self-reported frequent insomnia symptoms. We interpreted EDS alleles showing patterns of association with higher sleep efficiency, longer sleep duration, fewer discrete sleep bouts and fewer insomnia symptoms as reflective of greater sleep propensity; whereas EDS alleles associated with these sleep traits in a largely inverse manner were interpreted as reflective of disturbed sleep or a sleep fragmentation phenotype (Fig. 2). GRS of EDS loci stratified by the two clusters support our interpretation, with sleep propensity loci showing robust associations with early circadian traits, (e.g. morning chronotype P=1.22×10 -10 ; Table 1). Future statistically 1 0 robust clustering analysis approaches including other sleep and related traits will be needed to validate and distinguish the phenotypic subtypes of EDS 51

Functional effects of loci
The 42 loci lie in genomic regions encompassing 164 genes (Supplementary Table 12), and 3 associations are in strong linkage disequilibrium with known GWAS associations for other traits, including blood cell count, HDL cholesterol and caffeine metabolism [52][53][54][55] . Genes at multiple loci have been implicated in Mendelian syndromes or in experimental studies in mouse or fly models.
Eighteen loci harbor one or more genes with potential drug targets 56,57 .
We performed fine mapping analyses for potential causal variants using PICS 58 and identified 33 variants within 25 EDS loci with a causal probability larger than 0.2 (Supplementary Table   13). The majority of likely causal variants were intronic (65%) followed by non-coding transcript variants (8%) and nonsense mediated decay (NMD) transcript variants (7%) (Supplementary Fig. 5). Functional variants included a missense variant rs12140153 within PATJ, a synonymous variant rs11078398 within RAI1, regulatory variants rs10800796 in the promoter region of LMOD1, and rs239323 in a CTCF binding site in the gene POM121L2. Using the Oxford Brian Imaging Genetics (BIG) server 59 , we further observed the pleiotropic locus at rs13135092 (SLC39A8 60 ) to be significantly associated with bilateral putamen and striatum volume in the UK Biobank (p<2.8×10 -7 ; N=9,707; Supplementary Fig. 6). This could be of particular interest given the importance of these central brain centers in influencing motor and emotional behaviors, and emerging data implicating these centers in the integration of behavioral inputs that modulate arousal and sleep-wake states 61,62 .

Gene-based, pathway and tissue enrichment analyses
Gene-based analyses using PASCAL 63 identified 94 genes associated with EDS (p<2.29×10 -6 ) ( Supplementary Table 14 Genes at loci showing clustering with sleep propensity phenotypes (n=67) showed enriched expression in brain tissues including cortex, hippocampus, cerebellum, and amygdala (P<10 -3 ; Supplementary Fig. 7B). In contrast, no tissues were enriched in expression of genes that showed clustering with sleep fragmentation phenotypes (n=97) ( Supplementary Fig. 7C), suggesting that the mechanisms associated with sleep fragmentation may be more complex, reflective of multifactorial influences. Pathway and ontology analyses results for clustered genes using FUMA also reveal different patterns (Supplementary Fig. 8 and 9).
The heritability of EDS explained by genome-wide SNPs was estimated at 6.9% (SE=1%).
Partitioning heritability across tissue types and functional annotation classes indicated enrichment of heritability in central nervous system and adrenal/pancreas tissue lineage tissues, and in regions conserved in mammals, introns, and H3K4me1-potentially active and primed enhancers 69 (P<8.3×10 -4 ) (Supplementary Table 18).

Genetic overlap among EDS, sleep disorders and other disease traits
Consistent with EDS being a symptom of several sleep disorders, GRSs of genome-wide significant SNPs for restless leg syndrome 70 (P=0.0002), insomnia 71 (P=4×10 -7 ), and coffee consumption 54 (P=1.87×10 -12 ) (often used as a sleepiness "counter-measure") were significantly associated with EDS phenotype (Table 2). Although EDS is a key symptom of narcolepsy, the GRS of narcolepsy 45 was not associated with EDS phenotype (P=0.126), suggesting narcolepsy loci did not explain EDS variation in this sample. We could not examine the genetic overlap of sleep apnea loci and EDS because few significant loci for sleep apnea have been reported in the literature 35 and there was limited sleep apnea information in this cohort.
To investigate the genetic correlation between EDS and other common disorders, we tested the proportion of genetic variation of EDS shared with 233 other traits with published GWAS summary statistics in LDSC 72 . After adjusting for multiple comparisons, significant positive genetic correlations were observed for EDS with obesity traits, coronary heart disease, and psychiatric traits (P<0.0001) (Supplementary Table 19). The genetic correlations of EDS with coronary artery disease and psychiatric traits persisted after adjusting for BMI (Figure 2), perhaps partially reflecting shared neurologic or neuroendocrine factors, such as those that underlay insomnia and short sleep with cardiac and psychiatric traits 73,74 . Consistently suggestive negative genetic correlations for EDS with subjective well-being and reproductive traits (age at menarche and age at first birth) were also observed (P<0.005). Therefore, we performed sensitivity analysis using the Radial MR-Egger approach (Methods) 76 to control for bias due to pleiotropy, and observed an effect that was consistent with our main IVW analyses but less precisely estimated (wider confidence intervals) because this method is statistically relatively inefficient (β=0.025; 95% CI [-0.005,0.055]; P-value=0.103; Supplementary   Figure 1. Genome-wide association analysis of EDS (modelled as a 4-level continuous variable) identified 37 significant loci (P<5×10 -8 ), including previously associated signals near PATJ, HCRTR2, and CPEB1 (blue). Genes near novel significant loci are highlighted in green. 9 Figure 2. EDS risk alleles associated predominantly with sleep propensity or sleep fragmentation phenotypes. Each cell shows effect sizes (z-scores) of associations between sleep traits (accelerometry-derived sleep efficiency, sleep duration, number of sleep bouts, and self-reported insomnia symptoms) and EDS risk alleles (positively associated with EDS). Sleep propensity alleles were defined as more likely associated with higher sleep efficiency, longer sleep duration, fewer sleep bouts and fewer insomnia symptoms. Sleep fragmentation alleles were defined as more likely associated with lower sleep efficiency, shorter sleep duration, more sleep bouts and more insomnia symptoms. 0 on its ed ity n, as nd Figure 3. Top significant genetic correlations (r g ) between EDS and published summary statistics of independent traits using genomewide summary statistics using LD score regression (LDSC). Blue color indicates positive genetic correlation and red color indicates negative genetic correlation. Larger colored squares correspond to more significant P values, and asterisks indicate significant (P <2.2×10 -4 ) genetic correlations after adjusting for multiple comparisons of 224 available traits. All genetic correlations in this report can be found in tabular form in Supplementary Table 20.
1 Figure 4. Radial plot of two-sample Mendelian randomization between (A) BMI and (B) Type 2 diabetes with EDS outcome using IVW and MR-Egger tests. The x-axis is the inverse standard error (square root weights in the IVW analysis) for each SNP. The y-axis scale represents the ratio estimate for the causal effect of an exposure on outcome for each SNP ( ) multiplied by the same square root weight.

Population and study design
The discovery analysis was conducted on participants of European ancestry from the UK Biobank study 27 . The UK Biobank is a prospective study that has enrolled over 500,000 people aged 40-69 living in the United Kingdom. Baseline measures collected between 2006 -2010, including self-reported heath questionnaire and anthropometric assessments, were used in this analysis. Participants taking any self-reported sleep medication (described elsewhere 73 ) were excluded. 452,071 individuals of European ancestry were studied with available phenotypes and genotyping passing quality control, as described below.

Excessive daytime sleepiness and covariate measurements
Self-reported excessive daytime sleepiness (EDS) was ascertained in the UK Biobank using the question "How likely are you to dose off or fall asleep during the daytime when you don't mean to? (e.g. when working, reading or driving)" with the response options of "Never/rarely", "sometimes", "often", "all of the time", "do not know", and "prefer not to answer". Participants reporting "do not know" and "prefer not to answer" were set to missing. Other responses were coded continuously as 1 to 4 corresponding to the severity of EDS. The primary covariates used were self-reported age and sex, and body mass index (BMI) calculated as weight/height 2 .
Covariates used in the sensitivity analyses include potential confounders (depression, social economic status, alcohol intake frequency, smoking status, caffeine intake, employment status, marital status, and use of psychiatric medications) and indices of sleep disorders and sleep traits (daytime napping, sleep apnea, narcolepsy, sleep duration, insomnia, and chronotype).
Depression was recorded as a binary variable (yes/no) corresponding to question "Ever depressed for a whole week?". Social economic status was measured by the Townsend Deprivation Index based on aggregated data from national census output areas in the UK.
Alcohol intake frequency was coded as a continuous variable corresponding to "daily or almost daily", "three or four times a week", "once or twice a week", "once to three times a month", "special occasions only", and "never" drinking alcohol. Smoking status was categorized as "current', "past", or "never" smoked. Caffeine intake was coded continuously corresponding to self-reported cups of tea/coffee per day. Employment status was categorized as "employed", "retired", "looking after home and/or family", "unable to work because of sickness or disability", "unemployed", "doing unpaid or voluntary work", or "full or part-time student". Day napping was coded continuously ("never/rarely", "sometimes" or "usually") responding to the question "Do you have a nap during the day?" Sleep apnea cases (N=5,571) were identified as a union of self-reported and International Classification of Diseases (ICD)-10 coded (G47.3) sleep apnea.
Narcolepsy cases (N=7) were determined by the ICD-10 code (G47.4) Insomnia was recorded as "never/rare", "sometimes" or "usually" responding to the question "Do you have trouble falling asleep at night or do you wake up in the middle of the night?". Individuals reported "usually" were considered as frequent insomnia symptom cases. Sleep duration was recorded as discrete integers in response to the question "About how many hours sleep do you get in every 24 hours (please include naps)". In this study, short sleep was defined by sleep duration shorter than 7 hours and long sleep was defined by sleep duration longer than 8 hours. Chronotype was categorized as "definitely a 'evening' person", "more an 'evening' than a 'morning' person", "more a 'morning' than 'evening' person", and "definitely an 'morning' person". Secondary analyses were performed on participants further excluding shift workers, psychiatric mediation users, and participants with chronic and psychiatric illness (defined elsewhere 73 , N=255,426).
Briefly, for each individual, a 5-minute rolling median of the absolute change in z-angle (representing the dorsal-ventral direction when the wrist is in the anatomical position) across a 24-hour period. The 10th percentile of the output was used to construct an individual's threshold, distinguishing periods with movement from non-movement. Inactivity bouts were defined as inactivity of at least 30 minutes duration. Inactivity bouts with less than 60 minutes gaps were combined to blocks. The SPT-window was defined as the longest inactivity block, with sleep onset as the start of the block and waking time as the end of the block 85 . We applied exclusion criteria based on accelerometer data quality including 1) none-zero or missing in "data problem indicator" (Field 90002); 2) 0 in "good wear time" (Field 90015); 3) 0 in "good calibration" (Field 90016); 4) 0 in "calibrated on own data" (Field 90017); 5) "data recording errors" (Field 90182) > 788 (Q 3 +1.5×IQR); and 6) none-zero in "interrupted recording periods" (Filed 90180).
Accelerometry data from 85,502 participants of European ancestry passed quality control and were analyzed in this study.
The distributions of accelerometer data are described in Supplementary Table 1. The details of each measurement are as follows. L5 and M10 were the least-active five-hour window and mostactive ten-hour window for each day estimated from a moving average of a contiguous five/tenhour window. The L5 timing was defined as the number of hours elapsed from the previous midnight whereas M10 was defined as the number of hours elapsed from the previous midday.
Sleep midpoint was the midpoint between the start and end of the SPT-window. L5, M10, and sleep midpoint variables capture the circadian characteristics of an individual. Sleep episodes within the SPT-window were defined as periods of the z-axis angle change less than 5° for at least 5 minutes 83 . Sleep duration in a SPT-window was calculated as the sum of all sleep episodes. The mean and standard deviation of sleep duration across all SPT-windows were investigated in this study. Sleep efficiency was calculated as sleep duration divided the total SPT-window duration in a SPT-window. Sleep fragmentation was examined by counting the number of sleep episodes of at least 5 minutes separated by at least 5 seconds of wakefulness within a SPT-window. Diurnal inactivity duration was the total duration of estimated bouts of inactivity that fell outside of the SPT-window in 24 hours, which included both inactivity and naps. SNPs with minor allele frequency <0.001, genotype calling triplet probability < 0.1, or imputation quality <0.8 were further excluded from the current analyses. The detailed description of genotyping, QC, and imputation are available elsewhere 86 . We further performed K-means clustering using the principal components (PCs) of ~100,000 high quality genotyped SNPs (missingness <1.5% and MAF >2.5%) and identified 453,964 participants of European ancestry.

Genome-wide association analysis
We performed a genome-wide association analysis (GWAS) of self-reported EDS using 452,071 individuals of European ancestry in the UK Biobank across autosomes. A linear mixed regression model was applied adjusting for age, sex, genotyping array, 10 PCs and genetic relatedness matrix, using EDS as a continuous variable of 4 integers, using BOLT-LMM 28 .
Similar linear mixed regression analyses were performed additionally adjusting for BMI and stratified by sex. Secondary GWAS excluding related individuals, shift workers, individuals who used psychiatric medications, and participants with chronic health and psychiatric illness (N=255,426) was performed adjusting for age, sex, genotyping array and 10 PCs in PLINK 1.9 87 .
Trait heritability was estimated using BOLT-REML 28 . Genome-wide significance level was set at 5×10 -8 . Gene-sex and gene-health status interaction analyses were performed on unrelated individuals using a linear regression model in PLINK. Conditional analyses to dissect independent signals in significant genomic regions were performed using GCTA-COJO 88 .
Variant annotation was performed using PICS 58 .

Post-GWAS analyses Sensitivity and stratification analyses of significant loci
Sensitivity analyses of the genome-wide significant loci in the primary analysis (P<5×10 -8 ) were performed additionally adjusting for potential confounders (including depression, socio- hypersomnolence [defined as sleepiness plus long sleep duration without any chronic or psychiatric diseases], and 7-day accelerometry data). Linear or logistic regression analyses were performed adjusting for age, sex, genotyping array, and 10 PCs. Genome-wide summary statistics of sleep duration, insomnia, chronotype, long sleep duration, short sleep duration, and 7-day accelerometry using BOLT-LMM were available in public database 73,74,79,89 . We then performed cluster analysis on those loci according to their estimated effect sizes with objectively measured sleep duration, sleep efficiency, sleep fragmentation (number of sleep periods), and insomnia and interpreted their effects as sleep homeostasis or sleep fragmentation.

Gene, pathway and tissue-enrichment analyses
We further examined the genes within genome-wide significant loci using gene-based, pathway, and tissue enrichment analyses 63,67,68,90 . Gene-based analysis was performed using PASCAL 63 .
Pathway and ontology enrichment analyses were performed using FUMA 90 and EnrichR 67,68 .
Tissue enrichment analysis was performed using MAGMA 65 in FUMA, which controlled for gene size. Pathway and tissue enrichment analyses were also performed on genes within loci belonging to sleep propensity and sleep fragmentation clusters separately.
We constructed a weighted GRS comprised of the 42 significant EDS loci and tested for associations with other self-reported sleep traits (sleep duration, long sleep duration, short sleep duration, insomnia, chronotype, and day naps), and 7-day accelerometry traits in the UK Biobank 73,74,79,89 . Weighted GRS analyses were performed by summing the products or risk allele count multiplied by the effect estimate reported in the primary GWAS of EDS using R package gds (https://cran.r-project.org/web/packages/gds/gds.pdf). We also tested the GRSs of reported loci for insomnia, sleep duration, short sleep, long sleep, day naps, chronotype, restless leg syndrome (RLS), narcolepsy, and coffee consumption associated with EDS using the same 0 approach. The SNPs selected for each trait include 57 genome-wide significant loci for frequent insomnia 73 , 78, 27 and 8 loci for sleep duration, long sleep, and short sleep respectively 74 , 348 loci for chronotype 79 , 125 loci for daytime napping, 20 genome-wide significant loci for RLS 70 , 6 non-HLA suggestive significant loci (P<10 -4 ) in a narcolepsy case-control study of European Americans 43 , and 8 loci for coffee consumption 54 .

Genetic correlation analyses
Genetic correlation analysis using LD Score regression was performed on genome-wide SNPs mapped to the HapMap3 reference panel between EDS (with and without adjustment for BMI) and 233 published GWAS available in LDHub 72,91 . The significance level was determined as 10 -4 correcting for multiple comparisons. Pairwise genetic correlations among EDS, frequent insomnia, sleep duration, long sleep duration, short sleep duration and chronotype were performed locally using LDSC. We also partitioned heritability across 8 cell-type regions and 25 functional annotation categories available in LDSC 92 . Enrichment of the partitioning heritability was calculated in each region with and without extension (±500bp).

Mendelian randomization analyses
To investigate the causal relationship between EDS and other traits, we performed two-sample  [1984][1985][1986], HUNT2 [1995][1996][1997] and HUNT3 [2006][2007][2008]) have been completed including more than 120,000 individuals. EDS phenotype was collected in HUNT3 by asking the question "How often in the last 3 months have you felt sleepy during the day?" with the choices "Never/seldom", "Sometimes", and "Several times". Individuals with self-reported stroke, We studied exhaustion and fatigue in this population. This was ascertained by asking a question "During the past 30 days, have you felt yourself exhausted or overstrained?" with choises "Never", "Sometimes" and "Often". 20,344 individuals with both phenotype and whole genome genotyped and imputed data were available for this study. Genotyping was performed at the  I  d  e  n  t  i  f  i  c  a  t  i  o  n  o  f  n  o  v  e  l  r  i  s  k  l  o  c  i  f  o  r  r  e  s  t  l  e  s  s  l  e  g  s  s  y  n  d  r  o  m  e  i  n  g  e  n  o  m  e  -w  i  d  e  a  s  s  o  c  i  a  t  i  o  n  s  t  u  d  i  e  s  i  n  i  n  d  i  v  i  d  u  a  l  s  o  f  E  u  r  o  p  e  a  n  a  n  c  e  s  t  r  y  :  a  m  e  t  a  -a  n  a  l  y  s  i  s  .   L  a  n  c  e  t  N  e  u  r  o  l   1  6  ,  8  9  8  -9  0  7 ( 2 0 1 7  P  a  r  t  i  t  i  o  n  i  n  g  h  e  r  i  t  a  b  i  l  i  t  y  b  y  f  u  n  c  t  i  o  n  a  l  a  n  n  o  t  a  t  i  o  n  u  s  i  n  g  g  e  n  o  m  e  -w  i  d  e  a  s  s  o  c  i  a  t  i  o  n  s  u  m  m  a  r  y  s  t  a  t  i  s  t  i  c  s  .   N  a  t  G  e  n  e  t   4  7  ,  1  2  2  8  -3  5  (  2  0  1  5  )  .  9  3  .  B  u  r  g  e  s  s  ,  S  .  &  T  h  o  m  p  s  o  n  ,  S  .  G  .  B  i  a  s  i  n  c  a  u  s  a  l  e  s  t  i  m  a  t  e  s  f  r  o  m  M  e  n  d  e  l  i  a  n  r  a  n  d  o  m  i  z  a  t  i  o  n  s  t  u  d  i  e  s  w  i  t  h  w  e  a  k  i  n  s  t  r  u  m  e  n  t  s  .   S  t  a  t  M  e  d   3  0  ,  1  3  1  2  -2  3  (  2  0  1