Genome-wide association study of obstructive sleep apnoea in the Million Veteran Program uncovers genetic heterogeneity by sex

Abstract

Background: Genome-wide association studies (GWAS) for obstructive sleep apnoea (OSA) are limited due to the underdiagnosis of OSA, leading to misclassification of OSA, which consequently reduces statistical power. We performed a GWAS of OSA in the Million Veteran Program (MVP) of the U.S. Department of Veterans Affairs (VA) healthcare system, where OSA prevalence is close to its true population prevalence.

Methods: We performed GWAS of 568,576 MVP participants, stratified by biological sex and by harmonized race/ethnicity and genetic ancestry (HARE) groups of White, Black, Hispanic, and Asian individuals. We considered both BMI adjusted (BMI-adj) and unadjusted (BMI-unadj) models. We replicated associations in independent datasets, and analysed the heterogeneity of OSA genetic associations across HARE and sex groups. We finally performed a larger meta-analysis GWAS of MVP, FinnGen, and the MGB Biobank, totalling 916,696 individuals.

Findings: MVP participants are 91% male. OSA prevalence is 21%. In MVP there were 18 and 6 genome-wide significant loci in BMI-unadj and BMI-adj analyses, respectively, corresponding to 21 association regions. Of these, 17 were not previously reported in association with OSA, and 13 replicated in FinnGen (False Discovery Rate p-value < 0.05). There were widespread significant differences in genetic effects between men and women, but less so across HARE groups. Meta-analysis of MVP, FinnGen, and MGB biobank revealed 17 additional, previously unreported, genome-wide significant regions.

Interpretation: Sex differences in genetic associations with OSA are widespread, likely associated with multiple OSA risk factors. OSA shares genetic underpinnings with several sleep phenotypes, suggesting shared aetiology and causal pathways.

Funding: Described in acknowledgements.

Keywords: Electronic health records; Genome-wide association study; Obstructive sleep apnoea; Sex differences.

Fig. 1

Flowchart of the performed genetic analyses of OSA. The primary analysis is the multi-population OSA GWAS in MVP (orange colour box). This multi-population GWAS was a meta-analysis of sex- and HARE-group stratified analyses (top row of boxes). Various analyses were performed using the primary results, shown in the arrow coming down of the orange box. MVP analyses within sex strata and within Black and White HARE groups were used for assessment of potential group differences (two left most boxes in the bottom row). The MVP OSA GWAS summary statistics were also meta-analysed with GWAS summary statistics from MGB Biobank and FinnGen, followed by replication and proxy-replication look up in the Estonian Biobank and in open-access datasets (blue boxes). OSA: obstructive sleep apnoea. MVP: million veteran program. BMI: body mass index. BMI-adj: BMI-adjusted. BMI-unadj: BMI-unadjusted. HARE: harmonized race/ethnicity and genetic ancestry. MGB: Mass General Brigham.

Fig. 2

Proportions of OSA and BMI distribution across MVP HARE groups. The figure provides three panels. The top left panel visualizes the number of MVP study participants by HARE group and by biological sex. The bottom left panel visualizes the percentages of MVP participants who have OSA according to the EHR-based phenotype, stratified by HARE group and by sex. The right panel provides density plots visualizing the distributions of participants' BMI, stratified by HARE and sex. OSA: obstructive sleep apnoea. MVP: million veteran program. HARE: harmonized race/ethnicity and genetic ancestry. BMI: body mass index. EHR: electronic health records.

Fig. 3

Miami plots of the main MVP analyses and of the MVP, FinnGen, and MGB Biobank meta-analyses. Panel (a) provides a Miami plot from the primary MVP GWAS and panel (b) provides a Miami plot from the meta-analysis of MVP, FinnGen, and MGB Biobank. In each Miami plot, the top part provides results from BMI-unadj analysis, and the bottom from BMI-adj analysis. Each point corresponds to a SNP, with position on the x-axis determined by chromosomal location (genome build hg37) and position on the y-axis corresponding to the negative of the base-10 log of the association p-value of that SNP in association with OSA and in the respective analysis. Thus, sets of poin that appear to be vertically aligned correspond to sets of SNPs that are likely in linkage-disequilibrium with each other within the same genomic region. These vertically aligned points are annotated by the nearest gene. P-values were computed based on the 1-degree of freedom Wald test. MVP: million veteran program. GWAS: genome-wide association study. MGB: Mass General Brigham. SNP: single nucleotide polymorphism. BMI: body mass index. BMI-unadj: BMI-unadjusted. BMI-adj: BMI-adjusted.

Fig. 4

Association of OSA GRSs with OSA in the MGB Biobank. The figure provides estimated ORs and accompanying 95% confidence intervals of OSA per 1 standard SD increase in OSA GRSs constructed based on BMI-adjusted GWAS (“BMI-adj GRS”) and based on BMI-unadjusted GWAS (“BMI-unadj GRS”). The GWAS used to guide the construction of the GRSs are the primary, multi-population MVP GWAS, and were constructed as weighted sums of all SNPs reported in Table 1. Thus, SNP alleles were weighted by the estimated log-OR in MVP (Table 1). The GRSs were constructed, and association analyses were performed in the MGB biobank. Association analyses in MGB biobank were performed using logistic regression while adjusting to BMI (BMI-adj MGB analysis; left) and without adjusting to BMI (BMI-unadj MGB analysis; right). All MGB analyses were adjusted to age, sex, genetic batch, and the first 10 PCs of genetic data. OR: odds ratio. OSA: obstructive sleep apnoea. SD: standard deviation. GRS: genetic risk score. BMI: body mass index. BMI-adj: BMI-adjusted. GWAS: genome-wide association study. MVP: million veteran program. SNP: single nucleotide polymorphism. MGB: Mass General Brigham. PC: principal component.

Fig. 5

Protective genetic associations with OSA in females in SNPs with evident sex interactions. The figure compares the estimated OR of SNPs in association with OSA in male and in female participants from MVP. Each square corresponds to a set of SNPs defined by their p-value (p) in test of difference in their effect sizes between biological sex groups. SNPs with sex-difference effect p-values > 0.5 are not shown. Depicted results are from BMI-adj analysis. The blue line is the line of identity. The test statistics used to compute p-values are provided in the main manuscript. P-values were two-sided p-values computed assuming that the test statistics has standard normal distribution. OR: odds ratio. SNP: single nucleotide polymorphism. OSA: obstructive sleep apnoea. MVP: million veteran program. P: p-value. BMI: body mass index. BMI-adj: BMI-adjusted.

Fig. 6

Genetic correlations of OSA with cardiometabolic traits are driven by BMI. Visualization of the estimated genetic correlations between OSA and other phenotypes, in BMI-adj and BMI-unadj analyses. For each pair of phenotypes, genetic correlation was computed using LDSC based on summary statistics from GWAS, restricted to HapMap SNPs. Phenotypes are described in Supplemental Note S5. ∗ indicates LDSC p-value < 0.05. OSA: obstructive sleep apnoea. BMI: body mass index. BMI-adj: BMI-adjusted. BMI-unadj: BMI-unadjusted. LDSC: Linkage disequilibrium score regression. GWAS: genome-wide association study. SNP: single nucleotide polymorphism. DBP: diastolic blood pressure; SBP: systolic blood pressure; HTN: hypertension; LDL: low-density lipoprotein cholesterol; HDL: high-density lipoprotein cholesterol; TC: total cholesterol; TG: triglycerides; T2D: type 2 diabetes; AD: Alzheimer's disease. HapMap SNPs: tag SNPs identified by the haplotype mapping (HapMap) project as uniquely able to identify haplotype blocks across multi-ancestry populations.