Genome-wide association study of thyroid-stimulating hormone highlights new genes, pathways and associations with thyroid disease

Abstract

Thyroid hormones play a critical role in regulation of multiple physiological functions and thyroid dysfunction is associated with substantial morbidity. Here, we use electronic health records to undertake a genome-wide association study of thyroid-stimulating hormone (TSH) levels, with a total sample size of 247,107. We identify 158 novel genetic associations, more than doubling the number of known associations with TSH, and implicate 112 putative causal genes, of which 76 are not previously implicated. A polygenic score for TSH is associated with TSH levels in African, South Asian, East Asian, Middle Eastern and admixed American ancestries, and associated with hypothyroidism and other thyroid disease in South Asians. In Europeans, the TSH polygenic score is associated with thyroid disease, including thyroid cancer and age-of-onset of hypothyroidism and hyperthyroidism. We develop pathway-specific genetic risk scores for TSH levels and use these in phenome-wide association studies to identify potential consequences of pathway perturbation. Together, these findings demonstrate the potential utility of genetic associations to inform future therapeutics and risk prediction for thyroid diseases.

Fig. 1. Overview of the study design.

Flow diagram summarising the two-stage study design. UKB UK Biobank, EUR European ancestry, SA South Asian ancestry, SNP single nucleotide polymorphism, GRS genetic risk score, EstBB Estonian Biobank, G&H Genes & Health.

Fig. 2. 112 genes prioritised by two or more variant-to-gene criteria.

The first seven columns indicate that at least one variant implicates the corresponding gene via the evidence for that column (Supplementary Data 3). The remaining six columns indicate the strength of association of the most significant variant implicating the corresponding gene as causal with respect to the TSH increasing allele, such that shades of blue represent associations with the other thyroid phenotypes that have the same direction of effect as the TSH association and shades of red represent an opposite direction of effect to the TSH association (Supplementary Data 5). As there were no significant pQTL associations, that column has been omitted from the figure.

Fig. 3. PheWAS for select pathway-specific TSH-weighted GRS.

PheWAS for pathway-specific TSH-weighted GRS partitioned by: (a, top) activin receptor-like kinase (ALK) in cardiac myocytes pathway (Biocarta); (b, bottom) platelet activation, signalling and aggregation pathway (Reactome).

Fig. 4. PGS associations with TSH/free T4 across non-European ancestries in UK Biobank.

The TSH PGS association with TSH and free T4 across ancestry groups in UK Biobank shown as standard deviation (SD) change in TSH/free T4 per SD increase in the PGS. The ancestry groups were as defined by the Pan-UK Biobank initiative – AFR African ancestry, AMR admixed American ancestry, CSA Central/South Asian ancestry, EAS East Asian ancestry, MID Middle Eastern ancestry. Error bars indicate 95% confidence intervals. Sample size for TSH: AFR, n = 1430; AMR, n = 249; CSA, n = 3033; EAS, n = 721; MID, n = 479. Sample size for free T4: AFR, n = 584; AMR, n = 145; CSA, n = 1307; EAS, n = 362; MID, n = 256.

Fig. 5. Association of TSH PGS deciles with thyroid diseases.

The TSH PGS decile analysis for four clinical thyroid phenotypes—(a, top left) hypothyroidism, (b, top right) hyperthyroidism, (c, bottom left) thyroid cancer, and (d, bottom right) other thyroid disease. Statistical tests were two-sided, the height of the bars show the point estimate of the effect and whiskers show the 95% CI. OR odds ratio. The Mann-Kendall test is a test for monotonic trend. Sample size: Hypothyroidism, 29,550 cases and 368,691 controls; Hyperthyroidism, 5549 cases and 392,692 controls; Thyroid cancer, 624 cases and 358,144 controls; Other thyroid disease, 1233 cases and 124,515 controls.

Fig. 6. Association of TSH PGS with age-of-onset of hypothyroidism and hyperthyroidism.

Proportion of hypothyroidism (a, left) and hyperthyroidism (b, right) cases diagnosed by age stratified into lowest (grey), median (blue) and highest (yellow) decile for the TSH PGS. Shaded bands indicate 95% confidence intervals. Sample size: Hypothyroidism, 29,550 cases and 368,691 controls; Hyperthyroidism, 5549 cases and 392,692 controls. Statistical tests were two-sided.