ARTICLE

https://doi.org/10.1038/s41467-020-14625-1 OPEN

Insights into the aetiology of snoring from

observational and genetic investigations
in the UK Biobank
Adrián I. Campos 1,2,6, Luis M. García-Marín1,3,6, Enda M. Byrne4, Nicholas G. Martin 1,

Gabriel Cuéllar-Partida2,5,7* & Miguel E. Rentería 1,2,7*

1234567890():,;

Although snoring is common in the general population, its aetiology has been largely
understudied. Here we report a genetic study on snoring (n ~ 408,000; snorers ~ 152,000)
using data from the UK Biobank. We identify 42 genome-wide significant loci, with an SNP-
based heritability estimate of ~10% on the liability scale. Genetic correlations with body mass
index, alcohol intake, smoking, schizophrenia, anorexia nervosa and neuroticism are
observed. Gene-based associations identify 173 genes, including DLEU7, MSRB3 and POC5,
highlighting genes expressed in the brain, cerebellum, lungs, blood and oesophagus. We use
polygenic scores (PGS) to predict recent snoring and probable obstructive sleep apnoea
(OSA) in an independent Australian sample (n ~ 8000). Mendelian randomization analyses
suggest a potential causal relationship between high BMI and snoring. Altogether, our results
uncover insights into the aetiology of snoring as a complex sleep-related trait and its role in
health and disease beyond it being a cardinal symptom of OSA.

1 Genetic Epidemiology Lab, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia. 2 Faculty of Medicine, The University of Queensland,

Brisbane, QLD, Australia. 3 Tecnológico de Monterrey, Escuela de Ingeniería y Ciencias, Zapopan, Jalisco, México. 4 Institute for Molecular Bioscience, The
University of Queensland, Brisbane, QLD 4072, Australia. 5 University of Queensland Diamantina Institute, Brisbane, QLD, Australia. 6These authors
contributed equally: Adrián I. Campos, Luis M. García-Marín. 7These authors jointly supervised this work: Gabriel Cuéllar-Partida, Miguel E. Rentería.
*email: [email protected]; [email protected]

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

S
noring is the vibration of the upper airway structures that 95% confidence interval (CI) 1.009–1.012]) and, to a greater
occurs during sleep and creates noise as the air passes in extent, with sex (ORmales = 2.264 [2.212–2.316]). The prevalence
and out while breathing. Habitual snoring is common in the of sleep apnoea was higher within the snorer group (Table 1).
population, its overall prevalence increases with age and is higher Furthermore, BMI, SES, smoking frequency and alcohol con-
in males (35–45%) than females (15–28%)1. Importantly, snoring sumption frequency were also associated with snoring (Fig. 1a).
is a hallmark of obstructive sleep apnoea (OSA), a sleep-related Although snoring prevalence was higher in males, BMI was
breathing disorder characterized by repeated episodes of complete positively correlated with snoring prevalence in both males and
or partial obstructions of the upper airway during sleep, despite females (Fig. 1b). A lower SES, as determined by both Townsend
the effort to breathe1. OSA is usually associated with a reduction deprivation index and average household income, was associated
in blood oxygen saturation and is often accompanied by asso- with increased snoring in males only. Smoking frequency was
ciated daytime symptoms, such as excessive daytime sleepiness, positively correlated with snoring prevalence in females and, to a
fatigue and decreased cognitive function. Although the vast lesser extent, in males (Fig. 1a–c). In contrast, alcohol con-
majority of patients with OSA exhibit snoring, a minority sumption frequency was correlated with snoring in males and, to
(20–25%) of patients with central sleep apnoea do not snore2 and a lesser extent, in females (Fig. 1a–d). We further identified other
it is estimated that sleep apnoea may occur in as many as 20–40% factors such as whole-body fat mass and sleep duration that are
of the adult population that are snorers, leaving the remaining correlated with snoring (Supplementary Table 1).
60–80% of snorers in the category of habitual non-apnoeic benign
snorers. Snoring has previously been associated with body mass Discovery GWAS and SNP heritability. We performed a GWAS
index (BMI)3,4 as well as with the risk of cardiovascular disease study of snoring, taken as a dichotomous variable (n = 408,317;
such as coronary heart disease and stroke among postmenopausal cases ~152,000; controls ~256,000). After quality control (QC; see
women5. Twin and family studies have demonstrated the exis- Methods), 11,010,159 genetic variants remained in the analysis.
tence of a genetic predisposition to habitual snoring, with herit- This uncovered 127 independent genome-wide significant asso-
ability estimates suggesting that 18–28% of variance can be ciations across 41 genomic risk loci (Fig. 2a and Supplementary
accounted for by genetic factors4,6. A proportion of its heritability Fig. 1)13. Annotation for the top 15 risk loci is shown in Table 2
may be mediated through other heritable lifestyle factors such as and a list of all genomic risk loci is given in Supplementary
smoking and alcohol consumption, which can also contribute to Data 1. The overall SNP heritability on the liability scale (h2SNP)
snoring7–9. was 9.9% (SE = 0.39%).
Snoring is known to reduce sleep quality for both snorers and
their sleeping partners10,11, reducing energy/vitality and increas-
ing daytime anxiety12, risk of depression, stress, fatigue and Genetic correlations. The trait that showed the highest genetic
sleepiness11. Here we leverage data from the UK Biobank and an correlation with habitual snoring was self-reported sleep apnoea
Australian sample of adults, in an effort to characterize the (rG = 0.78, SE = 0.17, p-value = 3 × 10−05[ldsc χ2-test]; Supple-
molecular underpinnings of habitual snoring as a complex, mentary Data 2). We also analysed the genetic correlation
polygenic trait. We estimate the prevalence of snoring at 36% and between snoring and three measures of overnight oxyhemoglobin
identify phenotypic correlations with BMI, socio-economic status saturation: average SpO2, minimum SpO2 saturation and percent
(SES), smoking and alcohol consumption frequency. A genome- of sleep with oxyhemoglobin saturation under 90% (Perc90)14.
wide association study (GWAS) identifies 42 genome-wide sig- Minimum SpO2 and Perc90, which are known proxies for sleep-
nificant loci and a significant single-nucleotide polymorphism disordered breathing, but not average SpO2 (which reflects
(SNP)-based heritability of ~10%. We perform sensitivity ana- changes in ventilation not necessarily related to sleep apnoea),
lyses, which suggest that the genetic aetiology of snoring is not showed moderate significant genetic correlations with snoring
uniquely driven by BMI genetic factors. We further assess the (Fig. 3). Other traits genetically correlated to snoring included
existence of sex-specific effects, identifying two loci with evidence BMI, whole-body fat mass, sodium in urine, mood swings, cor-
of differential effect sizes, albeit in the same direction, between onary artery disease, alcohol intake frequency, pulse rate, current
males and females. No large-scale sex-specific genetic effects are tobacco smoking, heart disease, lung cancer, the ratio between
identified. We also employ statistical genetics methods such as forced expiratory volume in 1 s (FEV1) and forced vital capacity
linkage disequilibrium (LD) score regression and Mendelian (FVC), neuroticism, subjective wellbeing and heart rate, among
randomization (MR), to identify genetically correlated traits with others. Traits showing a negative genetic correlation with snoring
snoring and assess causality. Sleep-related traits such as sleep included schizophrenia, FVC, FEV1, fluid intelligence score,
apnoea, BMI, cardiometabolic and psychiatric traits are geneti- educational attainment, age at menarche, mean accumbens
cally correlated with snoring. MR analyses suggest a potential volume and anorexia nervosa. Overall, traits related to BMI, risk
causal relationship between higher BMI and snoring. Further- for psychiatric disease, lung function and heart disease were
more, we use our GWAS results to estimate individual polygenic among those with the strongest evidence of association (Fig. 3
scores (PGS) and predict snoring and probable OSA in an and Supplementary Data 2). Notably, pulse rate, whole-body fat
independent sample of Australian adults, highlighting the possi- mass and BMI were also phenotypically associated with snoring
bility of studying this complex sleep disorder using snoring as a in this sample (see above and Supplementary Table 1).
proxy phenotype.
Sensitivity analysis. We performed two follow-up sensitivity
GWAS to explore the effects of BMI, adiposity and nonlinear
Results effects on associated variants. The first sensitivity GWAS included
Snoring prevalence and risk factors. Our population-based BMI as a covariate, whereas the second included BMI, BMI2,
discovery sample consisted of 408,317 individuals of white British age × sex, age2 and whole-body fat mass. Both sensitivity analyses
ancestry from the UK Biobank. Participants in the sample were showed very similar results, with a genetic correlation of 0.9998
deemed as snoring ‘cases’ (37%) based on their report that a (SE = 0.0002). We therefore focus below on the simple model
partner or housemate had complained to the participant about adjusting only for BMI and basic covariates (see Methods). The
their snoring (see Methods and Table 1). Snoring was sig- results revealed 97 genome-wide significant SNPs across 34
nificantly associated with age (odds ratio (OR) = 1.011 [per year, genomic risk loci (Fig. 2a) with overall SNP heritability on the

2 NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1 ARTICLE

Table 1 Sample composition and descriptive statistics of UK Biobank discovery sample.

Female N (%) Apnoea N (%) Age mean (SD) BMI mean (SD)
Cases (snorers) 63833 (40.74%) 4510 (2.88%) 57.01 (7.70) 28.67 (4.85)
Controls 161775 (61.44%) 1663 (0.63%) 56.60 (8.21) 26.64 (4.52)
Total 225608 (53.72%) 6173 (1.47%) 56.75 (8.03) 27.39 (4.75)

N = Sample sizes. Descriptive statistics were calculated only for the subset of the data with European or British ancestry.

a b 0.7
Males
Females

Age 0.6

Snoring prevalence (95% c.i.)

BMI 0.5

0.4
Townsend deprivation index

0.3
Household income

0.2

Tobacco most days

Females Males
0.1
1 50 100
Tobacco occasionally BMI percentile

Alcohol very frequently

c d
0.50 0.50

Alcohol frequently
Snoring prevalence (95% c.i.)

Snoring prevalence (95% c.i.)

0.45
0.45

0.40
Alcohol occasionally 0.40

0.35
0.35
Alcohol rarely 0.30

0.30
0.25

1.0 1.2 1.4 1.6 1.8 No Occasionally Yes, on most days Never Rarely Occasionally Frequently Very frequently

Current tobacco smoking Frequency of alcohol consumption

Snoring OR

Fig. 1 Lifestyle and demographic factors associated with increased snoring. a Forest plot depicting the odds ratios of studied variables on snoring for
males (blue) and females (red). b Sex- and BMI-stratified prevalence of snoring. c Sex- and frequency of tobacco smoking-stratified snoring prevalence.
d Sex- and frequency of alcohol consumption-stratified snoring prevalence. Error bars denote the 95% confidence intervals estimated using the odds ratio
and SE (a) or 1000 pseudo replicates from a bootstrap resampling procedure (b, c, d).

liability scale (h2SNP) of 8.67% (SE = 0.39) (Table 3). Although genome-wide significant variants, we used positional and
most of the lead SNP effect sizes remained very similar, the signal expression quantitative trait loci (eQTL) mapping, as well as
in the FTO locus showed a strong change after BMI adjustment genome-wide gene-based association analyses. From positional
(Fig. 2b and Supplementary Data 1). Further, after adjusting for and eQTL mapping, we identified 149 protein-coding genes
BMI, snoring was no longer genetically correlated with BMI and mapping to a genome-wide significant SNP. The nearest genes to
adiposity-related phenotypes, nor with heart attack, hypertension, the top signals included DLEU7 on chromosome 13 and MSRB3
alcohol and smoking habits. Traits that remained genetically on chromosome 12. In addition to DLEU7 and MSRB3, other
correlated with snoring after adjusting for BMI included schizo- compelling genes (prioritized by positional or eQTL mapping) for
phrenia, educational attainment, sodium in urine and sleep- snoring included BCL11B, FTO, SMG6, ROBO2, NSUN3, SNAP91
related traits such as daytime dozing, sleep apnoea and excessive and BCL2, which have previously been associated with smok-
daytime sleepiness. Notably, measures of nocturnal hypoxia such ing15,16; BLC11B, FTO17, RNA5SP47117,18 and SND1 and NSUN3,
as minimum SpO2, average SpO2 and Perc90 showed small previously associated with alcohol consumption15,17–19; FTO and
increases in their genetic correlation with snoring after adjusting SND1, associated with coffee consumption20; LMO4 associated
for BMI (Fig. 3 and Supplementary Data 3). The genetic corre- with insomnia21; and RNA5SP471 with narcolepsy21,22. In addi-
lation between both the adjusted and unadjusted GWAS was high tion, ROBO2 was previously associated with chronotype23,24 and
(rG = 0.923, SE = 0.003, p-value = 1 × 10−300), suggesting that a multiple genes (DLEU7, MSRB3, FTO, ANAPC4, SMG6, SND1,
considerable amount of snoring predisposition is not fully SIM1, KCNQ5, CEP120, MACF1, SNAP91 and BCL2) previously
explained by BMI. associated with musculoskeletal traits such as height and heel
bone mineral density (Supplementary Data 1 and 2)25–28.
Positional, eQTL and gene-based test prioritization. To Genome-wide gene-based association analysis identified 179
gain insights into the functional consequences of individual genes associated with snoring beyond genome-wide significance

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

a 17.5
Snoring b 0.020 5
15.0

12.5 0.015
–log10(p value)

10.0
4
7.5 0.010

H0 = snoring – snoring bmi adj = 0

5.0

Effect size on snoring

2.5 0.005
3

–log10(p value)
BMI adjusted
0.0 FTO
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 0.000
0.0

–2.5 2
–0.005
–5.0
log10(p value)

–7.5 –0.010
–10.0 1

–12.5 –0.015

–15.0

–17.5 –0.020
–0.020 –0.015 –0.010 –0.005 0.000 0.005 0.010 0.015 0.020
Snoring adjusted for BMI Effect size on snoring

Fig. 2 Genetic variants associated with snoring with and without adjustment for BMI. a Results from the genome-wide association studies are presented
as a Miami (mirrored Manhattan) plot. The x-axis contains a point for each genetic variant passing QC and is ordered by chromosome and base position.
The distance between each variant and the x-axis is a function of the significance (p-value) of the association. For the top panel the −log10(p-value) is
graphed on the y-axis, whereas the log10(p-value) is shown for the bottom panel (BOLT-LMM mixed model association test p-value). The red dotted line
denotes the genome-wide significance threshold (p < 5e − 8). b Effect sizes of independent lead SNPs across snoring (x-axis) and snoring adjusted for BMI
(y-axis). The only locus (FTO) with a statistically significant difference is labelled.

Table 2 Top 15 genomic risk loci for snoring showing the top SNP for each locus.

SNP Chr Position NEA EA MAF Nearest gene gwasP β SE

rs592333 13 51,340,315 G A 0.4423 DLEU7 1.00E − 17 −0.00906 0.001051
rs10878269 12 65,791,463 T C 0.3499 MSRB3 2.30E − 16 −0.00886 0.001086
rs61597598 2 156,996,626 A G 0.1163 AC073551.1 5.10E − 15 −0.01189 0.001529
rs2307111 5 75,003,678 C T 0.3956 POC5 4.80E − 13 0.007667 0.00107
rs2664299 14 99,742,187 C T 0.4145 BCL11B 1.10E − 12 0.007503 0.001061
rs13251292 8 71,474,355 G A 0.4145 TRAM1 4.30E − 12 −0.00737 0.001067
rs57222984 17 43,758,898 G A 0.2654 CRHR1:RP11-105N13.4 (ncRNA) 5.40E − 12 −0.00843 0.00122
rs725861 10 9,063,776 G A 0.1918 RP11-42L9.2 1.00E − 11 −0.00908 0.001338
rs12119849 1 96,878,072 A G 0.0825 UBE2WP1 (pseudogene) 4.10E − 11 −0.01226 0.00186
rs796856741 16 53,799,278 GT G 0.4433 FTO 4.70E − 11 −0.00696 0.001059
rs12429765 13 40,745,860 G A 0.493 LINC00332 6.20E − 11 0.0068 0.001051
rs34811474 4 25,408,838 A G 0.2167 ANAPC4 1.30E − 10 0.007996 0.001237
rs7829639 8 78,215,352 G A 0.2972 AC105242.1 (miRNA) 1.40E − 10 −0.00741 0.001155
rs180107 17 67,930,772 T A 0.3698 AC002539.2 (miRNA) 2.10E − 10 −0.0068 0.00106
rs11409890 17 46,269,542 TA T 0.4821 SKAP1:RP11-456D7 2.20E − 10 0.006664 0.001061

(p < 2.636e − 6; Bonferroni-corrected threshold for 18,971 tested Fig. 3a). Genes expressed in blood vessel and tibial artery tissue
genes) several of which were consistent with the mapped genes. were associated with snoring, even after adjusting for BMI
After adjusting for BMI, 104 protein-coding genes were identified (Supplementary Fig. 3b). Given these associations, and an
mapping to a genome-wide significant SNP from the positional observed genetic correlation between snoring and pulse rate
and eQTL mapping, whereas 120 genes remained significantly (rG = 0.106, SE = 0.03, p-value = 0.001), we conducted a two-
associated with snoring, including both MSRB3 and DLEU7 (see sample generalized summary-data-based MR (GSMR)29 to test
Supplementary Fig. 2 and Supplementary Data 2 and 3). eQTL for a possible causal relationship. The analysis suggested a one-
data obtained from Genotype-Tissue Expression (GTEx) and way causal relationship in which snoring increased pulse rate
mapped with Functional Mapping and Annotation of Genome- (Supplementary Fig. 4). We further explored the association
Wide Association Studies (FUMA) highlighted significant SNPs between snoring and BMI, whole-body fat mass, blood pressure,
that were associated with the expression of genes in several tissues major coronary heart disease and heart attack. GSMR results
including the lungs, blood, oesophagus, breast mammary, tibial suggested a bidirectional causal relationship, with snoring
nerve, and several areas of the brain, such as the cerebellum and exerting a causal effect on BMI, but also BMI exerting a causal
hippocampus (Supplementary Fig. 3 and Supplementary Data 2). effect on snoring, and a similar pattern was observed for heart
In summary, many of the mapped genes for snoring have been attack. In addition, one-way causal relationships were seen for
previously associated with other traits and diseases, primarily whole-body fat mass causing snoring and for snoring causing
grouped into cardiometabolic, cognitive/neurological, respiratory an increase in blood pressure (Supplementary Fig. 4). To
and psychiatric (Fig. 4 and Supplementary Data 1). control for possible confounding due to sample overlap, we
To further assess whether significant genes converged in conducted GSMR analyses using sex-stratified GWAS results (see
functional gene sets and pathways, we conducted gene-set Methods). The results supported causal relationships between
enrichment analyses of tissue expression data (Supplementary BMI (and whole-body fat mass) causing snoring, whereas all

4 NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1 ARTICLE

Snoring
Sleep disorders Snoring adj BMI
Perc90
Daytime dozing
Excessive daytime sleepiness
avgSpO2
minSpO2
FEV1/FVC
FEV1
FVC
Mood swings
Neuroticism score
Happiness
Schizophrenia
Anorexia Nervosa
Stopped smoking for 6+ months
Cigarettes per day
Smoking/smokers in household
Current tobacco smoking
Alcohol intake frequency
Fluid intelligence score
Intelligence
Heart attack
High blood pressure
Hypertension Sleep
Coronary artery disease Respiratory
Diastolic blood pressure Psychiatric
Pulse rate Lifestyle
Heart rate Cognitive
Body mass index Cardiometabolic
Whole body fat mass Anthropometric

Body fat percentage

Waist-to-hip ratio

–0.6 –0.4 –0.2 0.0 0.2 0.4 0.6

rG

Fig. 3 Snoring is genetically correlated with lifestyle, psychiatric and respiratory traits. LD score-based estimates of the genetic correlation between
snoring (circles) or snoring adjusted for BMI (squares) and other complex traits (y-axis). All of the depicted traits had a statistically significant genetic
correlation with snoring after multiple testing correction (Benjamini–Hochberg; FDR < 0.05). Error bars show the SE of the genetic correlations. Not all
tested associations are shown due to lack of space; complete results are available in Supplementary Data 2.

median analyses supported a causal effect of BMI (and whole-

Table 3 SNP-based heritability of snoring on the body fat mass) on snoring. Notably, although the MR-Egger
liability scale. estimates did not reach statistical significance, the Egger intercept
was not significantly different from zero.
Trait h2SNP SE λGC Intercept
Snoring 9.9% 0.39% 1.428 1.04 (0.01)
Snoring adj. for BMI 8.67% 0.39% 1.368 1.03 (0.009) Sex-stratified GWAS. Given the higher prevalence of snoring in
Snoring males 8.77% 0.54% 1.200 1.01 (0.007) males, we conducted GWAS analyses stratified by sex. These
Snoring females 12.42% 0.57% 1.254 1.02 (0.007) analyses identified 4 and 25 genome-wide significant SNPs for
Snoring adj. for BMI males 7.72% 0.56% 1.200 1.01(0.008) snoring in males and females, respectively. SNP heritability on the
Snoring adj. for BMI females 10.85% 0.54% 1.253 1.02 (0.007)
liability scale (h2SNP) was 8.77% (SE = 0.54%) and 12.42% (SE =
LD score regression derived SNP-based heritability results. Estimates were transformed to the 0.57%), respectively, for males and females (Table 3). In the
liability scale, assuming equal population and sample prevalence. λGC is the genomic inflation
factor and intercept is the LD score regression intercept.
sensitivity analyses, SNP heritability (h2SNP) was slightly lower
after adjusting for BMI in both males 7.72% (SE = 0.56%) and
females 10.85% (SE = 0.54%) (Table 3). We identified two loci
(lead SNPs rs199797821 and rs200391180) with a significantly
other associations did not reach statistical significance after different effect size between sexes, although in the same direction.
controlling for multiple testing (Supplementary Table 2). We The cross-sex genetic correlation was high (rG = 0.914, SE =
performed further sensitivity analyses using five different MR 0.033, p-value = 1.91 × 10−160), and effect sizes and directions for
methods that test different assumptions (Fig. 5). The results of top hits were highly consistent in both the male and female
GSMR, inverse variance weighted (IVW-MR) and weighted samples (Supplementary Data 1 and Supplementary Fig. 5).

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 5

6
–log10(adjusted P -value)

0
2
4
6
8
10
a

Idiopathic pulmonary fibrosis

Chronic obstructive pulmonary disease or high blood pressure (pleiotropy)
Lung function (FVC)
ARTICLE

Chronic obstructive pulmonary disease or resting heart rate (pleiotropy)

Lung function (FEV1)
Parkinson’s disease
Generalized epilepsy

Intelligence
Subcortical brain region volumes
Intracranial volume
Hippocampal tail volume (corrected for total hippocampal volume)
Cognitive function

Neurological/neurodevelopmental or cognitive function Mathematical ability

FEV1/FVC ratio
Hippocampal volume

General neurological/neurodevelopmental or cognitive ability

Alzheimer’s disease in APOE e4- carriers
Autism spectrum disorder

Forced vital capacity

Multiple system atrophy
Subiculum volume

Forced expiratory volume

Hippocampal subfield CA3 volume

Response to bronchodilator
Hippocampal subfield CA1 volume
Alcohol use disorder (total score)
General factor of neuroticism
Neuroticism
Neurociticism

Chronic obstructive pulmonary disease

Anxiety/tension (special factor of neuroticism)
Risk-taking tendency (4-domain principal component model)
Smoking initiation
TRAM1

Worry too long after an embarrassing experience

or cognitive
RNA5SP471

Feeling hurt
Neurological/

Irritable mood
neurodevelopmental

Body mass index

Atrial fibrillation
Aortic root size
Hip circumference
Pulse pressure
MRSB3
SNAP91

Body fat distribution (arm fat ratio)

Type 2 diabetes
BMI in non-smokers
Craniofacial microsomia
Body mass index (joint analysis main effects and smoking interaction)
BMI (adjusted for smoking behaviour)
Childhood body mass index
SND1

SMG6
NSUN3

KCNQ5

ROBO2
BCL11B
ANKRD10

Red cell distribution width

Psychiatric

ANAPC4
UBE2WP1

Body mass index in physically inactive individuals

Respiratory/lung

Weight

Trait (GWAS catalog)

Obesity
Body mass index (SNP x SNP interaction)
FTO

Heel bone mineral density

SIM1

Waist circumference
DLEU7
RNU6-929P

BMI in smokers
C17orf67

Body mass index (joint analysis main effects and physical activity interaction)
CEP120

Body mass index (adult)

QRS duration
Red blood cell count
Systolic blood pressure
VKORC1

Body mass index in physically active individuals

Handedness (Right-handed vs. non-right-handed)
Male-pattern baldness
Cardiometabolic

MACF1

Handedness (Left-handed vs. non-left-handed)

BCL2

Reaction time
RNA5SP471 PTGES3

Sense of smell
Primary tooth development (number of teeth)
Sex hormone-binding globulin levels
Immunoglobulin light chain (AL) amyloidosis
Neuroticism

Hand grip strength

Schizophrenia

IgA nephropathy
Bipolar disorder

BMI

Headache
Anorexia nervosa

Stroke

Spherical equivalent or myopia (age of diagnosis)

Unipolar depression

Body fat
Risk-taking behaviour

Pulse rate

Response to metformin (lC50)

Other

Erythrocyte cadmium concentration in never smokers

Pulse pressure

Orofacial clefts
Atrial fibrillation

Type II diabetes

Respiratory

Myopia
Blood protein levels

Chin dimples
Myocardial infarction

Cardiometabolic

Cleft lip with or without cleft palate

Systolic blood pressure
Diastolic blood pressure
Coronary artery disease

Virologic severity in herpes simplex virus type 2 infection

Thyroid stimulating hormone levels
Macular thickness
Non-obstructive artery disease

Psychiatric/substance use

enrichment analysis (hypergeometric test) based on all prioritized genes against gene sets defined by traits in the GWAS catalogue.
Skin, hair and eye pigmentation (multivariate analysis)

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications

Triglycerides
Metabolite levels (5-HIAA)
Neurological/neurodevelopment or cognitive
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

Fig. 4 Snoring genes associated with other traits or diseases. a Venn diagram showing the nearest genes to the lead significant SNP per genomic risk loci
identified for snoring, categorized according to previously reported association with other traits or diseases in the GWAS catalogue. b Significant gene-set
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1 ARTICLE

Weighted mode

Weighted median

Inverse variance weighted

GSMR

MR Egger

0.00 0.01 0.02 0.03

MR effect size
BMI on snoring

Fig. 5 Assessing the causal relationship between BMI and snoring. Forest plot showing Mendelian randomization results testing for causal relationships
between BMI and snoring. Blue markers represent the estimate between the exposure effect sizes (female-only GWAS) and assessing its causal
relationship on snoring (male-only GWAS). Yellow markers show the exposure effect sizes (male-only GWAS) assessing its causal relationship on snoring
(female-only GWAS). Diamonds represent the effect size and error bars represent the 95% confidence intervals.

PGS and prediction on an independent sample. We used the consumption and snoring in females but did so in males34. Our
discovery GWAS summary statistics to derive PGS in an inde- results strengthen the evidence pointing to alcohol as a risk factor
pendent target sample from the Australian Genetics of Depres- for snoring and sleep apnoea, potentially through a weakening
sion Study (AGDS)78. The prevalence of self-reported recent (relaxation) effect in the jaw and pharyngeal muscles35. Our study
snoring was 32%, with a higher prevalence in males than in revealed that lower SES was associated with higher snoring pre-
females (43.2% and 28.1%, respectively). PGS for snoring were valence in males only. Considering that the analysis simulta-
significantly associated with recent snoring for all but one (p < = neously accounted for the effects of BMI, alcohol and tobacco
5e − 8) of the p-value inclusion thresholds (Fig. 6 and Supple- consumption frequency, we speculate that the snoring–SES
mentary Fig. 6). Participants in the highest snoring PGS decile association might be mediated through factors that are associated
had around twice the odds of reporting recent snoring and with a lower SES and differ between males and females (e.g.,
choking or struggling for breath during sleep (i.e., probable OSA; work-related exposures). Nonetheless, whether SES is causally
sample prevalence = 8.2%) compared with those on the lowest associated with snoring in males remains to be assessed. This
decile (Fig. 6a). Furthermore, the PGS showed a stronger asso- would require well-powered genetic correlates of SES on inde-
ciation with increasing frequency of snoring severity (Fig. 6b). pendent samples. The differences in risk factor effect sizes
Finally, we showed that the snoring PGS explained a significant between males and females might contribute to the overall
amount of variance in recent snoring (Fig. 6c). observed sex difference in snoring prevalence. Future studies
should leverage statistical genetics methods such as polygenic
Discussion scoring or MR to further characterize the role of SES, smoking
This study advances our understanding of the aetiology and and alcohol-related phenotypes in snoring and OSA.
genetic architecture of snoring. Overall, snoring prevalence was The sex differences described above motivated us to perform
higher in males than in females, having a strong and positive sex-stratified GWAS. The larger sample size of the female sub-
correlation with BMI, tobacco smoking and alcohol consumption group conferred more power to detect genetic associations in our
in both sexes. The effects of BMI, smoking and alcohol con- analyses. Notably, we identified a higher snoring SNP-based
sumption have been previously reported in other studies7,30,31. In heritability in females than in males and two loci that displayed
our study, tobacco smoking displayed a stronger association with statistically significant different effect sizes between sexes.
snoring in females compared with males, a result consistent with Nonetheless, the observed high cross-sex genetic correlation and a
observations in an independent sample of ~15,000 Europeans high concordance in effect size and direction among top hits
published in 200431. Previous studies provided evidence that suggest that differences in sex-stratified GWAS might be due to
women might have a greater susceptibility to chronic obstructive power differences between the male and female subsamples rather
pulmonary disease after smoking32 and a greater bronchial than the existence of large-scale sex-specific genetic effects. Future
hyperresponsiveness after methacholine challenge33. These results studies should assess whether the loci with evidence of sex-
suggest that smoking might be associated with snoring through specific effects are mediating the differential effects of SES,
an increased inflammatory response and irritation of the airways, alcohol or tobacco consumption frequency between sexes.
thus having a larger effect size on females, compared with males. Top genes identified from gene-based test analysis for snoring
Conversely, our study indicates that the frequency of alcohol included DLEU7 and MSRB3. Previous reports have associated
consumption has a stronger influence on snoring in males DLEU7 with heel bone mineral density25,36, BMI37,38, height39,40,
compared with that in females. This is consistent with a previous cardiovascular diseases41, systolic blood pressure41 and pulmon-
study that failed to identify an association between alcohol ary function decline (FEV)42. The association between snoring

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

a b 1.5

Probable OSA
1.4
3.0 Recent snoring

1.3

per SD of snoring PGS

Snoring odds ratio
1.2

2.5 1.1

1.0

0.9

2.0
0.8

once a week

1–2 times

3–4 times

5–7 times
per week

per week

per week
OR

Less than
1.5 c 1.3E–14 1.3E–14
7.6E–14
2.0E–13
12
0.8 2.0E–11
10

Variance explained (%)

6.1E–10

–log10(p value)
Nagelkerke’s R2
0.6
8
1.0
0.4 6

2.4E–04 4
0.2
2

0.0 0
1 2 3 4 5 6 7 8 9 10
p ≤ 5e –08

p ≤ 1e –05

p ≤ 0.001

p ≤ 0.01

p ≤ 0.05

p ≤ 0.1

p ≤ 0.5

p ≤ 1.0
Snoring PGS decile

Fig. 6 Polygenic scores predict snoring and probable apnoea in an independent sample. a Forest plot showing the odds ratios (and 95% CI) by decile of
polygenic score (PGS) for snoring from the UK Biobank discovery sample (relative to the bottom decile = 1) for recent snoring and probable sleep apnoea
measured in an independent target sample of ~8000 unrelated Australian adults from the Australian Genetics of Depression Study (AGDS). b Snoring PGS
predicting different snoring frequencies reported in AGDS target sample. c Variance of recent snoring in AGDS explained by PGS calculated from UK
Biobank summary statistics. The x-axis represents the p-value threshold used for variant inclusion during genetic scoring; the y-axis represents the amount
of variance explained (change in Nagelkerke R2). The colour of each bar represents the significance of the association between the PGS and recent snoring
(−log10 p-value), while the exact p-value (Wald’s test) is shown above each bar.

genes and heel bone mineral density could be mediated by BMI genome-wide hits) that the GWAS for self-reported sleep apnoea
due to the association between BMI and bone density docu- has in the UK Biobank. Our analyses suggest that a GWAS for
mented previously43. MSRB3 plays a relevant role in protein and snoring captures a substantial portion of the genetic contribution
lipid metabolism pathways44, and has been associated with hip- to sleep apnoea, highlighting the importance of studying symp-
pocampal volume45–47, lung function41,48,49, Alzheimer’s dis- toms on a subclinical threshold, an approach that has already
ease50,51, brain injuries52, novelty seeking53, deafness54 and proven useful at understanding the heterogeneity of other com-
height41. These results could be consistent with the fact that plex traits such as depression and neuroticism59,60. Our study will
severe snoring may incur in nocturnal oxygen desaturation55, enable future efforts aimed at understanding the underlying
diminishing neuropsychological functions and, in some cases, genetic architecture of OSA using multivariate statistical genetic
resulting in tissue damage56 and contributing to impairment of approaches.
memory consolidation processes57. However, more research is We also observed moderate correlations with BMI, obesity and
needed to test this hypothesis. whole-body fat mass. Other relevant correlations included lung
Genetic correlations between snoring and a variety of traits and function, neurological, cardiovascular and psychiatric diseases,
diseases were identified. Sleep-related traits, among others, sur- and traits such as alcohol consumption frequency and smoking.
vived the sensitivity analysis adjusting for BMI. It is important to This is consistent with the observed phenotypic associations on
point out that adjustment for a heritable covariate can not only the first part of this study. The high genetic correlation between
reduce power for estimating genetic correlations or causality snoring and snoring adj. BMI (rG = 0.923, SE = 0.003, p-value =
estimates, but also introduce collider bias58. We have therefore 1 × 10−300) supports the idea that the genetic architecture of
only discussed genetic correlations that existed before adjusting snoring cannot be explained simply by BMI. Notably, the genetic
for BMI. The traits with the highest genetic correlation with correlations between snoring and diseases such as asthma and
snoring were sleep apnoea and sleep-disordered breathing phe- allergic rhinitis, which are considered risk factors for sleep-
notypes, which is consistent with loud snoring being a diagnostic disordered breathing61, do not reach statistical significance. This
criterion for OSA. This observation is also remarkable given the could imply that the association of atopic diseases and sleep-
small sample size (<2000 cases) and therefore reduced power (no disordered breathing is not mediated through genetics, but future

8 NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1 ARTICLE

studies should assess this in a more systematic manner. Our if any, sex-specific genetic observations (e.g., differences in SNP
results highlight the importance and utility of studying snoring, heritability) are due to true genetic effects rather than power
and unveil opportunities for understanding highly related sleep differences between the samples. Studying the relationship
traits and disorders, including OSA. between snoring and craniofacial phenotypes could provide
Our initial MR results using GSMR suggested a mutual causal important insights, given that these traits are likely to share a
relationship between BMI and snoring, and a similar pattern was common aetiology. Nonetheless, there is a limited number of
observed for heart attack, but only a one-way causal relationship available GWAS summary statistics of craniofacial structure
of whole-body fat mass causing snoring. We hypothesized BMI to phenotypes. Finally, the fact that PGS for snoring predicted less
be more heterogeneous and potentially more pleiotropic than than 1% of the variance on recent snoring suggest that the GWAS
whole-body fat mass. In fact, MR is known to be confounded by is still underpowered71. The heritability for snoring in twin stu-
pleiotropy62. Interestingly, a one-way causal relationship between dies is estimated in the range of 18–28%6, although some of the
snoring and pulse rate, which survived adjustment for BMI, was missing heritability for snoring may come from dominant genetic
identified. Nonetheless, this association did not reach statistical effects, it is likely that an increased power and studying rare
significance when accounting for sample overlap (i.e., sex- variants72 yield more powered genetic predictors.
stratified GSMR) or when using other MR methods that also In summary, we provide insights into the aetiology of snoring,
account for pleiotropy. The only causal associations that survived its risk factors and genetic underpinnings. Our observational
sample overlap and multiple testing correction were BMI or analyses showed a higher prevalence of snoring in males com-
whole-body fat mass causing snoring. Evidence for a causal pared with that in females, and effects of age, BMI, SES smoking
association was also observed from methods such as IVW-MR and alcohol consumption. In addition, tobacco smoking showed a
and weighted median MR in addition to GSMR. Using more higher effect on snoring prevalence in females compared with
stringent methods such as MR-Egger, the association did not that in males, alcohol consumption displayed a higher effect on
retain statistical significance. Nonetheless, it is known that MR- snoring prevalence in males compared with that in females and
Egger is a less powered method63. Furthermore, the MR-Egger SES seemed to be only associated with snoring in males. GWAS
intercept was not significantly different from zero, thus suggesting identified 127 genome-wide significant associations across 41
that the IVW-MR causality estimate is likely to be unbiased64. In genomic risk loci with h2SNP = 9.9%. We found two loci with
addition, GSMR removes pleiotropic instruments using a HEIDI differential sex effect sizes, but no evidence for large-scale sex-
outlier filter and should be unbiased by pleiotropy29. Overall, we specific genetic differences. We showed that most of the SNP
believe this to be a compelling evidence of a causal effect of BMI heritability identified is not simply due to BMI. We also found
on snoring, but caution should be taken given the results from evidence of a causal relationship from BMI or whole-body fat
MR-Egger and when extrapolating this observation to other mass to snoring. Evidence of genetic overlap between snoring and
related traits such as OSA. The lower number of instruments other cardiometabolic, respiratory, neurological and psychiatric
available for snoring as an exposure (Supplementary Table 2) traits was found. Finally, we used the GWAS summary statistics
makes it hard to assess whether the lack of significant results to derive individual PGS and predict both recent snoring and
using snoring as an exposure was due to a lack of power or due to probable OSA in an independent sample of Australian adults,
a lack of a true causal effects. Future efforts could leverage novel thus confirming the relevance of snoring as a sleep-related
statistical genetics methods that use all the GWAS results to test complex trait. Future studies should aim at leveraging powered
whether the associations observed could be explained by a GWAS on craniofacial structures, alcohol and tobacco beha-
causality rather than pleiotropy65. viours, to assess whether they are causal of snoring and to assess
Finally, we assessed the validity of our GWAS results by using the amount of shared genetic overlap between OSA and habitual
genetic scoring on an independent sample of Australian adults snoring, as the latter may serve to boost the power of obstructive
with data on recent snoring. Our successful prediction of snoring sleep apnoea genetic studies.
using PGS supports the external validity of our genetic association
results. Remarkably, we predicted probable OSA using a snoring- Methods
derived PGS. Thus, investigating the aetiology of snoring could Discovery sample and phenotypic information. Participants included in the
also help uncover the aetiology and genetic architecture of OSA, a present study were of European ancestry from the UK Biobank. Briefly, this
task that has proved to be difficult and challenging66. Future resource recruited participants between 2006 and 2010 to assess lifestyle, anthro-
pometric and health-related variables. Participants self-reported on sleep-related
efforts could assess the utility of snoring-derived PGS as an traits. Snoring was assessed as a single item (Field-ID: 1210): “Does your partner or
addition to the current battery of tests used to more accurately a close relative or friend complain about your snoring?” This question could be
diagnose OSA67, particularly given the issue of potential OSA answered with “Yes”, “No”, “Don’t know”, or “Prefer not to answer”. We excluded
underdiagnosis68,69. participants whose answers were “Don’t know” (n = 29,309) or “Prefer not to
answer” (n = 6854) from our analyses (Supplementary Table 3 shows the total
Our results highlight the utility of studying snoring and pro- sample size for each GWAS, including sensitivity and sex-stratified analyses). OSA
vide important insights into its aetiology and genetic architecture. cases were determined on the basis of either ICD-10 diagnosis code or self-report
However, some limitations must be acknowledged. Analyses used of sleep apnoea diagnosis in the UK Biobank.
self-reported snoring with the item relying on a partner or close
friend complaining about the participant’s snoring. Thus, the case Ethical regulations. The UK Biobank study was approved by the National Health
definition might be subject to participant-specific recall and Service National Research Ethics Service (ref. 11/NW/0382) and all participants
subjective biases. Nonetheless, we hypothesize that this limitation provided written informed consent to participate in the UK Biobank study.
Information about ethics oversight in the UK Biobank can be found at https://
might result in the inclusion of some cases as controls (i.e., www.ukbiobank.ac.uk/ethics/. Regarding the AGDS, all participants provided
snoring participants living alone) and therefore bias our results informed consent prior to participating in the study. This study and all ques-
towards the null rather than creating false positives. To avoid tionnaires used for AGDS were approved by the QIMR Berghofer Human Research
confounding due to population stratification, we only included Ethics Committee.
samples of European ancestry in our analyses. This is particularly
important, given reports of ethnic differences associated with Data extraction and statistical analyses. Raw data were extracted from the UK
Biobank under Application Number 25331. For a description of the field codes and
snoring prevalence70. Nonetheless, excluding other populations instances used, refer to Supplementary Table 4. Data were re-coded to remove
can limit the generalizability of these results outside the popula- missing data and uninformative responses (e.g., “I don’t know” or “I would rather
tions studied. As previously discussed, we cannot identify which, not answer”). Phenotype-derived estimates such as prevalence and associations

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

between variables were calculated using python. Libraries such as NumPy (https:// SNP inclusion on the PGS calculation. PGS were calculated by multiplying the
docs.scipy.org/doc/numpy/user/) and SciPy (https://docs.scipy.org/doc/) were used effect size of a given SNP by the imputed number of copies (using dosage prob-
for descriptive statistics and statsmodels (http://conference.scipy.org/proceedings/ abilities) of the effect allele present in an individual. Finally, the SNP dosage effects
scipy2010/pdfs/seabold.pdf) was used to build logistic regression models to assess were summed across all loci per individual. To assess the association between the
correlates of snoring and calculate ORs. Snoring prevalence stratified plots (e.g., PGS and snoring and probable OSA, we employed a logistic regression (python
Fig. 1) were performed using seaborn v0.9.0. CIs were calculated using a boot- statsmodels). The target sample for snoring was a subset (n = 9026) of the AGDS
strapping procedure generating 1000 pseudo replicates of the data. with data on recent snoring collected through the self-reported item: ‘During the
last month, on how many nights or days per week have you had or been told you
had loud snoring’. The item for probable OSA was: ‘During the last month, on how
Genetic association analyses and QC. Discovery GWAS. Analyses were per-
many nights or days per week have you had or been told your breathing stops or
formed for both the sex-stratified and the full sample using the BOLT-LMM
you choke or struggle for breath’. For both items, a positive response was con-
software tool. All GWAS performed were adjusted for age, sex, genotyping array
sidered from one to two times per week up to five to seven times per week and the
and the first 20 genetic principal components as fixed effects. BOLT-LMM
answer ‘Rarely, less than once a week’ was excluded. Only a subset (n ~ 8000) of
accounts for cryptic relatedness and population stratification. We used a post-
highly unrelated individuals (genetic relatedness < 0.05) were included in the
GWAS strict QC procedure corresponding to minor allele frequency (MAF ≥
analyses.
0.005) and imputation quality (≥0.60).
Sensitivity analyses. Given the strong correlation between snoring and BMI, we
carried out GWAS for snoring using BMI as a covariate (snoring adj. BMI) with Reporting summary. Further information on research design is available in
BOLT-LMM software with a QC corresponding to MAF ≥ 0.005 and imputation the Nature Research Reporting Summary linked to this article.
quality (≥0.60).
Data availability
Post-GWAS annotation and functional mapping. SNP annotation was con- The full GWAS summary statistics for this study will be made available through the
ducted using the FUMA platform. Risk loci are defined as up to 250 kb based on NHGRI-EBI GWAS Catalogue (https://www.ebi.ac.uk/gwas/downloads/summary-
the most right and left SNPs from each LD block. Gene-based tests were performed statistics). Individual level data for UK Biobank participants are available to eligible
using Multi-marker Analysis of GenoMic Annotation (MAGMA) as implemented researchers through the UK Biobank (www.biobank.ac.uk). Individual level AGDS data
on the FUMA platform, which provides aggregate association p-values based on all can be made available to academic collaborators with an appropriate Data Transfer
variants located within a gene and its regulatory region. We used the GWAS Agreement. Collaboration proposals can be directed to Professor Nick Martin (nick.
summary statistics to conduct a MAGMA73 analysis in the FUMA13 platform [email protected]).
(https://fuma.ctglab.nl/). This analysis includes a gene-based test to detect sig-
nificant SNPs associated with snoring. The prioritized genes based on positional
and eQTL mapping were further used to perform gene-set enrichment analysis Code availability
against the traits available in the GWAS catalogue. Furthermore, we used FUMA to Code used as part of the work presented in this manuscript is available from the authors
perform tissue-enrichment analysis, based on data from the GTEx project (https:// upon reasonable request.
gtexportal.org/home/documentationPage).
Received: 22 October 2019; Accepted: 22 January 2020;
Genetic correlation analyses. We performed genetic correlation analyses to
estimate genetic correlations between the discovery, sensitivity and sex-stratified
snoring GWAS summary statistics using LD score regression (LDSC) as imple-
mented in the Complex Trait Genomics Virtual Lab (CTG-VL, http://genoma.io).
Further, to uncover genetically correlated traits with snoring, genetic correlation
analyses using LDSC were performed on the platforms CTG-VL and LDHub
(http://ldsc.broadinstitute.org/ldhub/), which aggregate summary statistics for References
GWAS on hundreds of traits. 1. Main, C. et al. Surgical procedures and non-surgical devices for the
management of non-apnoeic snoring: a systematic review of clinical effects and
associated treatment costs. Health Technol. Assess. 13, 1–208 (2009). iii, xi–xiv.
Mendelian randomization. Mendelian randomization (MR) is a method in which 2. Bauters, F. A. et al. Phenotype and risk burden of sleep apnea. Hypertension.
genetic variants (e.g., SNPs) are used as instrumental variables to determine causal 74, 1052–1062 (2019).
relationships between traits, environmental exposures, biomarkers or disease out- 3. Ferini-Strambi, L. et al. Snoring in twins. Respir. Med. 89, 337–340 (1995).
comes74; to satisfy the conditions for MR, it is not required to identify the
4. Jennum, P., Hein, H. O., Suadicani, P., Sørensen, H. & Gyntelberg, F. Snoring,
actual causal variant, because a marker in LD with the causal variant can serve
family history, and genetic markers in men. The Copenhagen Male Study.
as a proxy instrument75. Moreover, to draw conclusions in regard with casual
Chest 107, 1289–1293 (1995).
effects, three relevant assumptions must be taken into consideration: (I) genetic
5. Sands, M. et al. Self-reported snoring and risk of cardiovascular disease among
variants must be associated with the exposure of interest; (II) genetic variants must
not be confounded; (III) genetic variants must be independent of the outcome postmenopausal women (from the Women’s Health Initiative). Am. J. Cardiol.
through other mechanisms76. We used GSMR29, an approach that leverages the 111, 540–546 (2013).
usage of multiple independent variables (SNPs) strongly associated with the out- 6. Carmelli, D., Bliwise, D. L., Swan, G. E. & Reed, T. Genetic factors in self-
come, to overcome these assumptions, as implemented in the CTG -VL (http:// reported snoring and excessive daytime sleepiness: a twin study. Am. J. Respir.
genoma.io). We used GSMR to assess causal relationships between snoring Crit. Care Med. 164, 949–952 (2001).
and BMI, whole-body fat mass and pulse rate using our results and existing 7. Katila, M., Saarenpää-Heikkilä, O., Saha, M.-T., Vuorela, N. & Paavonen, E. J.
summary statistics for these traits. To avoid possible confounding from sample Parental reports showed that snoring in infants at three and eight months
overlap, we performed GSMR using the summary statistics derived from sex- associated with snoring parents and smoking mothers. Acta Paediatr. 108,
stratified GWAS. For example, the female snoring GWAS results were used as 1686–1694 (2019).
exposure, whereas the male pulse rate GWAS results were used for the outcome. 8. Wei, Y. X. et al. [Epidemiological characteristics and correlated factors of
Finally, sensitivity two-sample MR analyses (MR-Egger, median-weighted esti- habitual snoring among Chinese aged 30 to 79 year-old]. Zhonghua Liu Xing
mator, IVW-MR and weighted mode) were performed using the R library Bing Xue Za Zhi 40, 917–923 (2019).
MRBase77 and UK Biobank sex-stratified summary statistics to ensure non- 9. Morales-Muñoz, I., Koskinen, S. & Partonen, T. The effects of seasonal
overlapping samples. affective disorder and alcohol abuse on sleep and snoring functions in a
population-based study in Finland. J. Sleep. Res. 27, e12611 (2018).
10. Blumen, M. et al. Effect of sleeping alone on sleep quality in female bed
Target sample and polygenic scoring. To quantify for the cumulative genetic
associations for snoring, we calculated PGS using a clumping + thresholding partners of snorers. Eur. Respir. J. 34, 1127–1131 (2009).
approach. Study description and sample characteristics of the AGDS are available 11. Beninati, W., Harris, C. D., Herold, D. L. & Shepard, J. W. Jr. The effect of
elsewhere78. Genotyping was conducted using the Illumina Infinium Global snoring and obstructive sleep apnea on the sleep quality of bed partners. Mayo
Screening Array platform and genotype imputation using the Haplotype Reference Clin. Proc. 74, 955–958 (1999).
Consortium’s reference panel in the Michigan Imputation Server79 was carried out 12. Doherty, L. S., Kiely, J. L., Lawless, G. & McNicholas, W. T. Impact of nasal
after performing standard QC procedures. Briefly, for PGS estimation, we excluded continuous positive airway pressure therapy on the quality of life of bed
indel, strand ambiguous- and low (R2 < 0.6) imputation quality variants. The most partners of patients with obstructive sleep apnea syndrome. Chest 124,
significant SNPs were selected using a conservative clumping procedure 2209–2214 (2003).
(PLINK1.980; p1 = 1, p2 = 1, r2 = 0.1, kb = 10,000) to correct for inflation arising 13. Watanabe, K., Taskesen, E., van Bochoven, A. & Posthuma, D. Functional
from LD. Eight PGS were calculated using different p-value thresholds (p < 5 × mapping and annotation of genetic associations with FUMA. Nat.
10−8, p < 1 × 10−5, p < 0.001, p < 0.01, p < 0.05, p < 0.1, p < 0.5, p < 1) as criteria for Commun. 8, 1826 (2017).

10 NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1 ARTICLE

14. Cade, B. E. et al. Associations of variants In the hexokinase 1 and interleukin 43. Salamat, M. R., Salamat, A. H., Abedi, I. & Janghorbani, M. Relationship
18 receptor regions with oxyhemoglobin saturation during sleep. PLoS Genet. between weight, body mass index, and bone mineral density in men referred
15, e1007739 (2019). for dual-energy X-ray absorptiometry scan in Isfahan, Iran. J. Osteoporos.
15. Liu, M. et al. Association studies of up to 1.2 million individuals yield new 2013, 205963 (2013).
insights into the genetic etiology of tobacco and alcohol use. Nat. Genet. 51, 44. Kim, H.-Y. & Gladyshev, V. N. Methionine sulfoxide reduction in mammals:
237–244 (2019). characterization of methionine-R-sulfoxide reductases. Mol. Biol. Cell 15,
16. Justice, A. E. et al. Genome-wide meta-analysis of 241,258 adults accounting 1055–1064 (2004).
for smoking behaviour identifies novel loci for obesity traits. Nat. Commun. 8, 45. van der Meer, D. et al. Brain scans from 21,297 individuals reveal the genetic
14977 (2017). architecture of hippocampal subfield volumes. Mol. Psychiatry (2018). https://
17. Kranzler, H. R. et al. Genome-wide association study of alcohol consumption doi.org/10.1038/s41380-018-0262-7
and use disorder in 274,424 individuals from multiple populations. Nat. 46. Hibar, D. P. et al. Novel genetic loci associated with hippocampal volume. Nat.
Commun. 10, 1499 (2019). Commun. 8, 13624 (2017).
18. Whitfield, J. B. et al. Biomarker and genomic risk factors for liver function test 47. Bis, J. C. et al. Common variants at 12q14 and 12q24 are associated with
abnormality in hazardous drinkers. Alcohol. Clin. Exp. Res. 43, 473–482 hippocampal volume. Nat. Genet. 44, 545–551 (2012).
(2019). 48. Wain, L. V. et al. Genome-wide association analyses for lung function and
19. Karlsson Linnér, R. et al. Genome-wide association analyses of risk tolerance chronic obstructive pulmonary disease identify new loci and potential
and risky behaviors in over 1 million individuals identify hundreds of loci and druggable targets. Nat. Genet. 49, 416–425 (2017).
shared genetic influences. Nat. Genet. 51, 245–257 (2019). 49. Shrine, N. et al. New genetic signals for lung function highlight pathways and
20. Zhong, V. W. et al. A genome-wide association study of bitter and sweet chronic obstructive pulmonary disease associations across multiple ancestries.
beverage consumption. Hum. Mol. Genet. 28, 2449–2457 (2019). Nat. Genet. 51, 481–493 (2019).
21. Lane, J. M. et al. Biological and clinical insights from genetics of insomnia 50. Gabbita, S. P., Aksenov, M. Y., Lovell, M. A. & Markesbery, W. R. Decrease in
symptoms. Nat. Genet. 51, 387–393 (2019). peptide methionine sulfoxide reductase in Alzheimer’s disease brain. J.
22. Luca, G. et al. Clinical, polysomnographic and genome-wide association Neurochem. 73, 1660–1666 (1999).
analyses of narcolepsy with cataplexy: a European Narcolepsy Network study. 51. Adams, S. L. et al. Methionine sulfoxide reductase-B3 (MsrB3) protein
J. Sleep. Res. 22, 482–495 (2013). associates with synaptic vesicles and its expression changes in the hippocampi
23. Jansen, P. R. et al. Genome-wide analysis of insomnia in 1,331,010 individuals of Alzheimer’s disease patients. J. Alzheimers Dis. 60, 43–56 (2017).
identifies new risk loci and functional pathways. Nat. Genet. 51, 394–403 52. Conner, S. C. et al. Methionine sulfoxide reductase-B3 risk allele implicated in
(2019). Alzheimer’s disease associates with increased odds for brain infarcts. J.
24. Jones, S. E. et al. Genome-wide association analyses of chronotype in 697,828 Alzheimers Dis. 68, 357–365 (2019).
individuals provides insights into circadian rhythms. Nat. Commun. 10, 343 53. Service, S. K. et al. A genome-wide meta-analysis of association studies of
(2019). Cloninger’s Temperament Scales. Transl. Psychiatry 2, e116 (2012).
25. Kim, S. K. Identification of 613 new loci associated with heel bone mineral 54. Ahmed, Z. M. et al. Functional null mutations of MSRB3 encoding
density and a polygenic risk score for bone mineral density, osteoporosis and methionine sulfoxide reductase are associated with human deafness DFNB74.
fracture. PLoS ONE 13, e0200785 (2018). Am. J. Hum. Genet. 88, 19–29 (2011).
26. Kim, K. et al. High-density genotyping of immune loci in Koreans and 55. Block, A. J., Hellard, D. W. & Cicale, M. J. Snoring, nocturnal hypoxemia, and
Europeans identifies eight new rheumatoid arthritis risk loci. Ann. Rheum. the effect of oxygen inhalation. Chest 92, 411–417 (1987).
Dis. 74, e13 (2015). 56. Kou, Z., Ye, Y. & Haacke, E. M. Evaluating the role of reduced oxygen
27. Morris, J. A. et al. An atlas of genetic influences on osteoporosis in humans saturation and vascular damage in traumatic brain injury using magnetic
and mice. Nat. Genet. 51, 258–266 (2019). resonance perfusion-weighted imaging and susceptibility-weighted imaging
28. Medina-Gomez, C. et al. Life-course genome-wide association study meta- and mapping. Top. Magn. Reson. Imaging 24, 253–265 (2015).
analysis of total body BMD and assessment of age-specific effects. Am. J. Hum. 57. Rasch, B. & Born, J. About sleep’s role in memory. Physiol. Rev. 93, 681–766
Genet. 102, 88–102 (2018). (2013).
29. Zhu, Z. et al. Causal associations between risk factors and common diseases 58. Aschard, Hugues et al. Adjusting for heritable covariates can bias effect
inferred from GWAS summary data. Nat. Commun. 9, 224 (2018). estimates in genome-wide association studies. Am. J. Hum. Genet. 96.2,
30. Nagayoshi, M. et al. Risk factors for snoring among Japanese men and 329–339 (2015).
women: a community-based cross-sectional study. Sleep. Breath. 15, 63–69 59. Thorp, J. G. et al. Genetic heterogeneity in self-reported depressive symptoms
(2011). identified through genetic analyses of the PHQ-9. Psychol. Med. 1–12 (2019)
31. Franklin, K. A. et al. The influence of active and passive smoking on habitual 60. Nagel, M., Watanabe, K., Stringer, S., Posthuma, D. & van der Sluis, S. Item-
snoring. Am. J. Respir. Crit. Care Med. 170, 799–803 (2004). level analyses reveal genetic heterogeneity in neuroticism. Nat. Commun. 9,
32. Jordan, R. E. et al. Sex, susceptibility to smoking and chronic obstructive 905 (2018).
pulmonary disease: the effect of different diagnostic criteria. Analysis of the 61. Koinis-Mitchell, Daphne et al. Sleep and allergic disease: a summary of the
Health Survey for England. Thorax 67, 600–605 (2012). literature and future directions for research. J. Allergy Clin. Immunol. 130.6,
33. Leynaert, B., Bousquet, J., Henry, C., Liard, R. & Neukirch, F. Is bronchial 1275–1281 (2012).
hyperresponsiveness more frequent in women than in men? A population- 62. Verbanck, M., Chen, C.-Y., Neale, B. & Do, R. Detection of widespread
based study. Am. J. Respir. Crit. Care Med. 156, 1413–1420 (1997). horizontal pleiotropy in causal relationships inferred from Mendelian
34. Peppard, P. E., Austin, D. & Brown, R. L. Association of alcohol consumption randomization between complex traits and diseases. Nat. Genet. 50, 693–698
and sleep disordered breathing in men and women. J. Clin. Sleep. Med. 3, (2018).
265–270 (2007). 63. Schmidt, A. F. & Dudbridge, F. Mendelian randomization with Egger
35. Issa, F. G. & Sullivan, C. E. Alcohol, snoring and sleep apnea. J. Neurol. pleiotropy correction and weakly informative Bayesian priors. Int. J.
Neurosurg. Psychiatry 45, 353–359 (1982). Epidemiol. 47.4, 1217–1228 (2017).
36. Kemp, J. P. et al. Identification of 153 new loci associated with heel bone 64. Burgess, Stephen & Simon, G. Thompson Interpreting findings from
mineral density and functional involvement of GPC6 in osteoporosis. Nat. Mendelian randomization using the MR-Egger method. Eur. J. Epidemiol.
Genet. 49, 1468–1475 (2017). 32.5, 377–389 (2017).
37. Shungin, D. et al. New genetic loci link adipose and insulin biology to body fat 65. O’Connor, L. J. & Price, A. L. Distinguishing genetic correlation from
distribution. Nature 518, 187–196 (2015). causation across 52 diseases and complex traits. Nat. Genet. 50, 1728–1734
38. Tachmazidou, I. et al. Whole-genome sequencing coupled to imputation (2018).
discovers genetic signals for anthropometric traits. Am. J. Hum. Genet. 100, 66. Pham, L. V. & Polotsky, V. Y. Genome-wide association studies in obstructive
865–884 (2017). sleep apnea. Will we catch a black cat in a dark room? Am. J. Respir. Crit. Care
39. Wood, A. R. et al. Defining the role of common variation in the genomic and Med. 194, 789–791 (2016).
biological architecture of adult human height. Nat. Genet. 46, 1173–1186 67. McNicholas, W. T. Diagnosis of obstructive sleep apnea in adults. Proc. Am.
(2014). Thorac. Soc. 5, 154–160 (2008).
40. Lango Allen, H. et al. Hundreds of variants clustered in genomic loci and 68. Costa, L. E. et al. Potential underdiagnosis of obstructive sleep apnoea in the
biological pathways affect human height. Nature 467, 832–838 (2010). cardiology outpatient setting. Heart 101, 1288–1292 (2015).
41. Kichaev, G. et al. Leveraging polygenic functional enrichment to improve 69. Motamedi, K. K., McClary, A. C. & Amedee, R. G. Obstructive sleep apnea: a
GWAS power. Am. J. Hum. Genet. 104, 65–75 (2019). growing problem. Ochsner J. 9, 149 (2009).
42. Imboden, M. et al. Genome-wide association study of lung function decline in 70. Goldstein, N. A. et al. Racial/ethnic differences in the prevalence of snoring
adults with and without asthma. J. Allergy Clin. Immunol. 129, 1218–1228 and sleep disordered breathing in young children. J. Clin. Sleep. Med. 7,
(2012). 163–171 (2011).

NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications 11

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14625-1

71. Dudbridge, F. Power and predictive accuracy of polygenic risk scores. PLoS Author contributions
Genet. 9, e1003348 (2013). M.E.R. and G.C.-P. conceived and directed the study. A.I.C. and L.M.G.-M. performed
72. Wainschtein, P. et al. Recovery of trait heritability from whole genome most of the statistical and bioinformatics analyses, with support from G.C.-P. and M.E.R.
sequence data. Preprint at bioRxiv 588020 (2019). https://doi.org/10.1101/ E.M.B. and N.G.M. collected and contributed data from the Australian sample. A.I.C.,
588020 L.M.G.-M., M.E.R. and G.C.-P. wrote the paper with feedback from all co-authors.
73. de Leeuw, C. A., Mooij, J. M., Heskes, T. & Posthuma, D. MAGMA:
generalized gene-set analysis of GWAS data. PLoS Comput. Biol. 11, e1004219
(2015).
Competing interests
The authors declare no competing interests.
74. Swerdlow, D. I. et al. Selecting instruments for Mendelian randomization in
the wake of genome-wide association studies. Int. J. Epidemiol. 45, 1600–1616
(2016). Additional information
75. Davey Smith, G. & Hemani, G. Mendelian randomization: genetic anchors for Supplementary information is available for this paper at https://doi.org/10.1038/s41467-
causal inference in epidemiological studies. Hum. Mol. Genet. 23, R89–R98 020-14625-1.
(2014).
76. Benn, M. & Nordestgaard, B. G. From genome-wide association studies to Correspondence and requests for materials should be addressed to G.C.-P. or M.E.R.
Mendelian randomization: novel opportunities for understanding
cardiovascular disease causality, pathogenesis, prevention, and treatment. Peer review information Nature Communications thanks the anonymous reviewer(s) for
Cardiovasc. Res. 114, 1192–1208 (2018). their contribution to the peer review of this work. Peer reviewer reports are available.
77. Gibran, H. et al. The MR-Base platform supports systematic causal inference
across the human phenome. eLife (2018). https://doi.org/10.7554/eLife.34408 Reprints and permission information is available at http://www.nature.com/reprints
78. Byrne, E. M. et al. The Australian Genetics of Depression Study: Study
Description and Sample Characteristics. Preprint at biorXiv (2019) https://doi. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
org/10.1101/626762 published maps and institutional affiliations.
79. Das, S. et al. Next-generation genotype imputation service and methods. Nat.
Genet. 48, 1284–1287 (2016).
80. Purcell, S. et al. PLINK: a tool set for whole-genome association and
Open Access This article is licensed under a Creative Commons
population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007).
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Acknowledgements Commons license, and indicate if changes were made. The images or other third party
This research was conducted using data from the UK Biobank resource under application material in this article are included in the article’s Creative Commons license, unless
number 25331. Data collection for the Australian Genetics of Depression Study was indicated otherwise in a credit line to the material. If material is not included in the
possible, thanks to funding from the Australian National Health & Medical Research article’s Creative Commons license and your intended use is not permitted by statutory
Council (NHMRC) to N.G.M. (GNT1086683). A.I.C. is supported by a UQ Research regulation or exceeds the permitted use, you will need to obtain permission directly from
Training Scholarship from The University of Queensland (UQ). M.E.R. thanks the sup- the copyright holder. To view a copy of this license, visit http://creativecommons.org/
port of the NHMRC and Australian Research Council (ARC), through a NHMRC-ARC licenses/by/4.0/.
Dementia Research Development Fellowship (GNT1102821). G.C.-P. is funded by an
ARC Discovery Early Career Researcher Award (DE180100976). We thank our colleague
Jackson Thorp for proof-reading our manuscript and useful feedback and discussion. © The Author(s) 2020

12 NATURE COMMUNICATIONS | (2020)11:817 | https://doi.org/10.1038/s41467-020-14625-1 | www.nature.com/naturecommunications