Genome-wide association study and polygenic risk prediction of hypothyroidism

Abstract

We performed a genome-wide meta-analysis of hypothyroidism (113,393 cases and 1,065,268 controls), free thyroxine (191,449 individuals) and thyroid-stimulating hormone (482,873 individuals). We identified 350 loci associated with hypothyroidism, including 179 not previously reported, 29 of which were linked through thyroid-stimulating hormone. We found that many hypothyroidism risk loci regulate blood cell counts and the circulating inflammasome, and through multiple gene-mapping strategies, we prioritized 259 putative causal genes enriched in immune-related functions. We developed a polygenic risk score (PRS) based on more than 115,000 hypothyroidism cases to address diagnostic challenges in individuals with or at risk of thyroid hormone deficiency. We show that the highest predictive accuracy for hypothyroidism was achieved when combining the PRS with thyroid hormones and thyroid-peroxidase autoantibodies, and that the PRS was able to stratify risk of progression among individuals with subclinical hypothyroidism. These findings demonstrate the potential for a hypothyroidism PRS to support the prediction of disease progression and onset in thyroid hormone deficiency.

Fig. 1. Hypothyroidism lead variants and their associations with thyroid hormones.

a, Relationships between minor allele frequencies and ORs for the 350 lead variants that were identified in the hypothyroidism genome-wide meta-analysis (113,393 cases and 1,065,268 controls) or through an endophenotype-driven analysis using thyroid-stimulating hormone genome-wide associations as priors. Coding variants are squared, new associations are turquoise, and known associations are gray. b, Relationships between hypothyroidism risk and changes in thyroid-stimulating hormone for 349 of 350 lead variants. c, Relation between hypothyroidism risk and change in free thyroxine for 348 of 350 lead variants. In b and c, the centerline represents the linear regression, and the shaded error band shows the 95% CI around the regression line. Statistical associations were assessed using two-sided Pearson correlation tests. No multiple testing correction was applied for these correlation analyses.

Fig. 2. PRS association with and prediction of hypothyroidism.

a, Associations between 10 deciles of the PRS and risk of hypothyroidism are presented as OR point estimates ± 95% CI error bars, estimated using logistic regression models adjusted for age, sex and PCs. No adjustments were made for multiple comparisons. b, Prediction of incident hypothyroidism cases. The benchmark model consisted of age, sex and four PCs. Prevalent risk factors for hypothyroidism were added iteratively to the benchmark model. The center of each error bar represents the AUC, and error bars indicate the 95% CIs, displayed in absolute terms on the right. No adjustments were made for multiple comparisons. MS, multiple sclerosis; PMR, polymyalgia rheumatica; PsA, psoriatic arthritis; RA, rheumatoid arthritis; DS, Down syndrome; SSc, systemic sclerosis; SLE, systemic lupus erythematosus; Celiac, celiac disease; Sjögren, Sjögren’s disease; T1D, type 1 diabetes.

Fig. 3. Progression from SCH to overt disease.

Ten-year cumulative incidence of disease progression from SCH to overt hypothyroidism in 8,114 primary care patients from the UKB. Lines represent the cumulative incidence, and shaded bands indicate the 95% CI. The green line represents individuals with low polygenic risk (<10th percentile), yellow represents intermediate polygenic risk (10th–90th percentile), and red represents high polygenic risk (>90th percentile). Cumulative incidence was estimated using the Aalen-Johansen estimator, which accounts for the competing risk of death. HRs with 95% CIs were estimated using two-sided Cox proportional hazards models, adjusted for age, sex and four PCs. No adjustments were made for multiple comparisons.

Fig. 4. Stratifying hypothyroidism risk using lifestyle characteristics and polygenic risk in the UKB.

a, Risk for incident hypothyroidism according to different lifestyle characteristics and categories. Data are presented as HR point estimates ±95% CIs, derived from two-sided Cox proportional hazards models, adjusted for age, sex and PCs. The center of each error bar represents the mean HR estimate. b,c, Ten-year risk of hypothyroidism, stratified by sex (b, females; c, males), age group, obesity status (BMI > 30), exercise regularity (yes/no), smoking status (yes/no) and divisions within the PRS across ten deciles (Q1–10).

Fig. 5. Phenome-wide associations between the hypothyroidism PRS and cancer and cardiometabolic phenotypes in the UKB.

The figure shows associations between the hypothyroidism PRS and 50 binary disease outcomes. OR reflect the change in disease risk per 1 s.d. increase in the PRS, estimated using logistic regression models adjusted for age, sex and four PCs. P-values were calculated using two-sided Wald tests. Each colored triangle indicates a significant association after Bonferroni correction (P < 0.001, that is, 0.05/50). Upward-pointing triangles indicate increased risk and downward-pointing triangles indicate decreased risk.