A Genetic Risk Score for Thyroid Peroxidase Antibodies Associates With Clinical Thyroid Disease in Community-Based Populations

Abstract

Context:

Antibodies against thyroid peroxidase (TPOAbs) are detected in 90% of all patients with Hashimoto thyroiditis, the most common cause of hypothyroidism. Hypothyroidism is associated with a range of adverse outcomes. The current knowledge of its genetic underpinnings is limited.

Objective:

The purpose of this study was to identify novel genetic variants associated with TPOAb concentrations and positivity using genome-wide association data and to characterize their association with thyroid function and disease.

Design, Setting, and Participants:

We studied European ancestry participants of 3 independent prospective population-based studies: Atherosclerosis Risk In Communities study (n = 7524), Study of Health in Pomerania (n = 3803), and Study of Health in Pomerania-TREND (n = 887).

Exposure:

Single nucleotide polymorphisms (SNPs), individually and combined into a genetic risk score (GRS), were examined.

Main Outcomes:

The main outcomes were TPOAb concentrations and positivity, thyroid hormone concentrations (TSH, free T₄), and clinical thyroid diseases (subclinical and overt hypothyroidism and goiter).

Results:

Significantly associated single nucleotide polymorphisms (P < 5 · 10⁻⁸) mapped into 4 genomic regions not previously implicated for TPOAbs (RERE, extended HLA region) and into 5 previously described loci. A higher Genetic Risk Score (GRS) based on these 9 SNPs showed strong and graded associations with higher TPOAb, TSH, and lower free T₄ concentrations (P < .001). Compared with individuals in the lowest GRS quartile, those in the highest quartile had 1.80-fold higher odds of subclinical hypothyroidism (95% confidence interval, 1.27–2.55) and 1.89-fold higher odds of overt hypothyroidism (95% confidence interval, 1.24–2.87).

Conclusion:

The identification of 4 novel genetic loci associated with TPOAb concentrations and positivity gives further insight into the genetic underpinnings of hypothyroidism. A GRS showed strong and graded associations with markers of thyroid function and disease in independent population–based studies.

Hypothyroidism has been related to fatigue, depression, heart failure, metabolic syndrome, and mortality (1–5). Thyroid dysfunction is seen in up to 10% of the adult population, and its prevalence increases with age (6). The most common cause of hypothyroidism in iodine-sufficient areas of the world is Hashimoto thyroiditis, which is characterized by gradual autoimmune-mediated destruction of the thyroid gland. High autoantibody titers against thyroid peroxidase (TPOAbs) are a sensitive clinical marker of Hashimoto thyroiditis and are detected in 90% of all patients with Hashimoto thyroiditis as opposed to 5% to 24% in the general population (6, 7). Despite the prevalence and adverse outcomes of autoimmune-mediated thyroid disease, its etiology remains incompletely understood (6, 8–11), complicating the identification of individuals at risk.

Autoimmune thyroid disease (AITD) is thought to arise from a combination of genetic susceptibility and environmental factors. Substantial progress has been made recently in the identification of such genetic susceptibility factors to AITD and other autoimmune diseases (12). Among the reported genetic risk loci are associations with HLA class I and II genes, CTLA-4, PTPN22, IL2RA, FOXP3, CD40, and FCRL3 (7, 9, 12). Estimates from twin studies indicate that the heritability to develop TPOAbs is around 70% (13), but the identified risk loci for AITD have been reported to account for only a minor proportion of the heritability (14, 15).

Genome-wide association studies (GWASs) are one way to gain novel insights into the pathophysiology of complex diseases. We undertook this study for several reasons: first, with use of previous GWAS findings as a basis, to identify additional novel genetic variants via meta-analysis with additional data from an independent study; second, to characterize associations of novel and known genetic variants in 3 large, community-based populations (the Atherosclerosis Risk in Communities [ARIC] study and the 2 Study of Health in Pomerania [SHIP] cohorts) with different levels of iodine supply; and third, to gain insight into the combined effect of the genetic variants by constructing a genetic risk score (GRS) and evaluating the GRS associations with measures of thyroid dysfunction and disease.

Subjects and Methods

Study populations

The ARIC study is a population-based prospective observational cohort of 15 792 adults in 4 US communities aged 45 to 64 years at the baseline visit in 1987 to 1989. Details of the study design were reported previously (16). In brief, 4 visits, each 3 years apart, were conducted between 1987 and 1998; a fifth visit was conducted from 2011 to 2013. Similar to previous large genetic studies of TPOAb concentrations (14), analyses in this report were limited to 7524 ARIC study participants of European ancestry with nonmissing information on thyroid hormone measurements, genotype information, and covariates.

The SHIP is a population-based project in northeastern Germany consisting of 2 independent longitudinal cohorts, SHIP (SHIP-0) and SHIP-TREND (SHIP-T) (17–19). For SHIP-0, 4308 participants aged 20 to 81 years were recruited from 1997 to 2001. For SHIP-T, 4420 Caucasian adults aged 20 to 81 years were recruited from 2008 to 2012. Written informed consent was obtained from all study participants for the ARIC and SHIP studies, and the protocol was approved by all ethic boards of the participating studies.

Measurement of thyroid hormones and ultrasound examination

Measures of thyroid function in the ARIC study were assayed in 2012 to 2013 from all participants with stored serum samples available from the second (1990–1992) and fifth (2011–2013) ARIC visits. Assays from Roche Diagnostics were implemented on an Elecsys 2010 analyzer (visit 2) and a Roche e411 analyzer (visit 5). All interassay coefficients of variation were ≤10% (Supplemental Table 1). Thyroid hormone intake was ascertained from information on the current intake of medications collected by trained interviewers.

For SHIP-0, serum TSH concentrations were analyzed on a Byk-Sangtec Diagnostica analyzer. The functional sensitivity of the assay was 0.03 mIU/L (17). In SHIP-T, serum TSH concentrations measured on a Siemens Dimension Vista analyzer. The functional sensitivity of the TSH assay was 0.014 mIU/L. Serum TPOAbs were measured on an Elias Medizintechnik GmbH VARELISA analyzer. The functional sensitivity of this assay was 1 IU/mL (Supplemental Table 1).

Thyroid ultrasonography was performed in SHIP-0 using an ultrasound VST-Gateway with a 5-MHz linear array transducer (Diasonic). In SHIP-T, ultrasonography was performed with a portable device using a 13-MHz linear array transducer (Vivid I; General Electric). In both cohorts, intra- and interobserver reliabilities were assessed before the baseline of the studies and semiannually during the studies. All measurements of the thyroid volume for interobserver and interdevice variability showed mean differences (2 SD) of the mean bias of <5% (<25%) (18). As described in the pertinent references, thyroid volume was calculated as length × width × depth × 0.479 (mL) for each lobe (20). Goiter was defined as a thyroid volume of >18 mL in women and >25 mL in men (21, 22). The normal thyroid echo pattern was classified as homogeneous. A homogeneous echo pattern with reduced echogenicity was defined as hypoechogenic.

Thyroid marker reference ranges and definition of thyroid disease

In clinical practice, the TPOAb status (positive or negative) is important in a first categorization of patients (6, 23), but both the presence and the concentrations of TPOAbs are relevant for disease onset (15, 24). Furthermore, different pathophysiological processes may be involved in the initiation and severity of the autoimmune response (15, 25). Besides TPOAb positivity, we therefore also tested the associations with continuous TPOAb concentrations. TPOAb positivity in the ARIC study was defined as >34 kIU/L as recommended by the assay manufacturer. Reference values for TSH, free T₄ (FT₄), and T₃ in the ARIC study were derived from their distributions in a healthy subset of ARIC participants (6). Participants were excluded from this healthy subset for the following reasons: ethnicity other than European ancestry or African American, self-reported goiter and/or thyroid disease before visit 2, intake of thyroid medication, estrogen, androgen, and/or lithium use, laboratory evidence of overt hyperthyroidism (TSH of <0.27 mIU/L; FT₄ of >21.88 pmol/L), overt hypothyroidism (TSH of >4.2 mIU/L; FT₄ of <11.97 pmol/L), elevated TPOAb concentrations (>34 kIU/L), or missing information on TSH, FT₄, T₃, or TPOAb measurements. Among the resulting 7394 participants, reference intervals were established using the 2.5th and 97.5th percentiles for thyroid hormone measurements. For European American (EA) individuals, reference intervals were as follows: TSH, 0.61 to 5.4 mIU/L; and FT₄, 11.2 to 19.3 pmol/L (Supplemental Table 2). Subclinical hypothyroidism was defined as elevated serum TSH concentrations but normal FT₄ concentrations and overt hypothyroidism as elevated TSH concentrations and FT₄ below the established reference range (3).

For SHIP-0 and SHIP-T, TPOAb positivity as well as the TSH and FT₄ reference values were defined (TPOAb positivity as >60 kIU/L for men and >100 kIU/L for women; SHIP-0: TSH, 0.3–3.0 mIU/L; and FT₄, 7.7–23.2 pmol/L; SHIP-T: TSH, 0.3–3.74 mIU/L; and FT₄, 8.4–18.9 pmol/L) as provided by the manufacturer and implemented previously (26). In SHIP-T, FT₄ was only measured for samples with TSH values outside of the reference range. Subclinical and overt hypothyroidism were defined as in the ARIC study using the SHIP-specific reference ranges.

In the ARIC study, 2 measures of thyroid function markers were available a median of 20.8 years (20.1, 21.5 years) apart, allowing us to evaluate prospective associations. Because of the long observational time gap between study visits 4 and 5 with limited information on potential informative censoring, incident thyroid disease associations were only evaluated in secondary analyses. Incident TPOAb positivity at ARIC visit 5 was defined as above among those who were TPOAb negative at the second visit. Incident subclinical and overt hypothyroidism at visit 5 were defined as above among participants with TSH and FT₄ concentrations within the reference range at visit 2.

Identification of associated genetic variants and genotyping information

To identify TPOAb-associated single nucleotide polymorphisms (SNPs) that could be combined into a GRS, we used information from a recently published large GWAS meta-analysis of TPOAb concentrations and positivity, to which the SHIP studies contributed but the ARIC study did not (14). The study reported 20 index SNPs associated with measures of AITD at P < 10⁻⁶ (7 associated with TPOAb concentrations, 10 with TPOAb positivity, and 3 with both traits). In the ARIC study, genotypes at these SNPs were available from the Affymetrix Genome-Wide Human SNP Array 6.0. Quality control procedures and genotype imputation were described previously (27). The published effect estimates for the 20 index SNPs were meta-analyzed with the corresponding SNP estimates and traits from the ARIC study using a sample-size weighted meta-analysis as implemented in METAL (28). Subsequently, genotypes of all previously identified replicated loci and newly identified genome-wide significant (P < 5 · 10⁻⁸) index SNPs were used for the calculation of the GRS. Genotyping and imputation quality of the SNPs studied in this report are listed in Supplemental Table 3.

GRS calculation and statistical analysis

The GRS for each individual was calculated across the associated SNPs as the weighted sum of all TPOAb positivity risk alleles carried by a person, with weights proportional to the effect estimates from the previously published meta-analysis (14). GRS quartiles were generated from the global distribution of the score. Supplemental Table 4 shows handling of participants with thyroid hormone measurements above the upper limit of detection and below the lower limit of detection before association analyses (14). We then estimated the age-, sex- and study center–adjusted associations between each of the SNPs and the natural logarithm of TPOAb concentrations as well as inverse normal transformations of TSH and FT₄ concentrations. Associations with thyroid volume, goiter, and low echogenicity were examined using multiple linear and logistic regression as appropriate. The proportion of TPOAb variance explained was estimated from a linear regression model.

The odds ratio (OR) of TPOAb positivity was estimated for each GRS quartile compared with the lowest GRS quartile (reference). The P value for trend was estimated by entering the GRS as a continuous variable into the regression model. The associations between the GRS quartiles and transformed thyroid hormone concentrations, as well as subclinical and overt hypothyroidism, thyroid volume, goiter, and low echogenicity were computed in an analogous fashion. Analyses on subclinical and overt hyperthyroidism were not conducted because of the very small case numbers. In a secondary analysis, associations of the GRS with incident TPOAb positivity and subclinical hypothyroidism were estimated using a multivariable adjusted Poisson regression model.

All GRS analyses were conducted using Stata/SE 13 or the R statistical framework (version 2.15.3; R Project for Statistical Computing; www.r-project.org).

Results

Study population and thyroid marker characteristics

Table 1 shows TPOAb concentrations and positivity in all cohorts. The prevalence of TPOAb positivity was 16% among 7524 ARIC participants, 7% among 3803 SHIP-0 participants, and 4% among 887 SHIP-T participants. Thyroid hormone replacement therapy ranged from 5.4% (ARIC) to 10% (SHIP-T), and the prevalence of goiter in both SHIP studies was around 35%.

Table 1.

Characteristics of the Study Populations

	ARIC	SHIP-0	SHIP-T
Sample size	7524	3803	887
Female sex, % (n)	55 (4101)	48 (1844)	54 (475)
Age, y, mean ± SD	57 ± 5.7	49 ± 16.3	50 ± 13.7
TSH, mIU/L, median (p25, p75)	1.9 (1.3, 2.9)	0.66 (0.44, 0.97)	1.2 (0.81, 1.6)
FT4, ng/dL, mean ± SD	1.1 ± 0.19	1.0 ± 0.3	NA
TPOAb, IU/mL, median (p25, p75)	10.9 (7.9, 18.0)	4.7 (1.9, 34.4)	11.0 (10.0, 14.0)
TPOAb positivity, % (n)	16 (1240)	7 (265)	4 (36)
Thyroid hormone replacement, % (n)	5.4 (407)	6.7 (272)a	10.0 (99)a
Antithyroid drug use, % (n)	0.03 (2)	NA	NA
Goiter, % (n)	NA	38 (1428)	34 (303)

Abbreviation: NA, not available.

^aGenotyped samples before exclusion.

Nine genetic loci associated with TPOAbs

Table 2 shows a summary of the SNP with the lowest P value (index SNP) at 9 genetic loci that contained variants associated with TPOAb concentrations (n = 1), positivity (n = 3), and both (n = 5) at genome-wide significance. Four of these 9 common SNPs, rs301799 near RERE and 3 markers in the HLA region (rs3094228, rs1894407, and rs9277555), mapped into a region that showed genome-wide significant associations with TPOAb measures for the first time and thus represent novel risk loci for elevated TPOAb concentrations and positivity.

Table 2.

GWSAs With TPOAb Concentrations or TPOAb Positivity

Variant	Chr	Position (hg 19)	Genesa	Risk Allele/Other Allele	Risk Allele Frequency	TPOAb Concentrationsb		TPOAb Positivityb
						Z Score	P Value	Z Score	P Value
rs301799	1	8489302	RERE	C/T	0.43	4.22	2.5E–05	5.68	1.4E–08
rs1230666	1	114173410	MAGI3, [PTPN22]	A/G	0.14	6.65	2.9E–11	5.52	3.5E–08
rs11675434	2	1407815	TPO	T/C	0.42	8.67	4.3E–18	9.57	1.1E–21
rs2010099	3	124300257	KALRN	C/T	0.89	4.92	8.8E–07	3.73	1.9E–04
rs3094228	6	31429927	HCP5, MICA	C/T	0.19	5.14	2.7E–07	5.69	1.3E–08
rs1894407	6	32787036	HLA-DOB, TAP2	C/A	0.61	5.49	4.0E–08	5.67	1.4E–08
rs9277555	6	33055605	HLA-DPB1	G/A	0.74	6.94	3.9E–12	6.30	3.0E–10
rs10944479	6	90880393	BACH2	A/G	0.19	4.89	1.0E–06	6.19	5.8E–10
rs653178	12	112007756	ATXN2, [SH2B3]	C/T	0.50	6.28	3.5E–10	7.10	1.3E–12

^aAnnotation is based on RefSeq genes, hg19: variant in gene if underlined, closest gene, or flanking genes if both are in same linkage disequilibrium block. Nearby genes that contain correlated functional variants are enclosed in brackets.

^bNumber of individuals with available SNP information (n) ranges from 30501–35332. Sample size varies because of missing genotype information across studies and between markers because of handling of values outside the assay detection range.

Individual SNP association with serum thyroid function tests

Next, the 9 associated SNPs were characterized for their association with TPOAb concentrations, and additional markers of thyroid function and disease in the ARIC and SHIP studies. Supplemental Table 5 shows that all risk alleles showed positive associations with TPOAb concentrations in the combined ARIC and SHIP studies, all of which except rs2010099 were statistically significant after correction for multiple testing (P < 5.6 · 10⁻³). The SNPs rs11675434 (TPO) and rs9277555 (HLA region) showed nominally significant associations (P < .05) with all 3 thyroid function measures evaluated (TPOAbs, TSH, and FT₄).

In the individual studies, the direction of effect for the risk allele at each SNP was almost all positive and of comparable magnitude across studies for TPOAb concentrations and TSH concentrations and mostly negative for FT4 concentrations, indicating that the risk alleles not only associate with higher TPOAb concentrations but also with higher TSH and lower FT4 concentrations.

The GRS associates with thyroid function markers and thyroid dysfunction

In the ARIC study, the proportion of variance explained in TPOAb concentrations increased from 3.2% using a GRS based on the 5 previously published loci to 4.1% by the addition of the novel 4 risk loci identified here. Using this improved GRS, we studied the effect of the continuous GRS on TPOAb, TSH, and FT₄ concentrations, as well as on the odds of TPOAb positivity, elevated TSH, and overt hypothyroidism. Figure 1 shows a strong and graded positive association of the GRS with TPOAb concentrations (Figure 1A) and TSH concentrations (Figure 1C) and a strong, graded, inverse association with FT₄ (Figure 1E). The predicted probability of TPOAb positivity ranged from 5% to 35% across the GRS range (Figure 1B). This effect was also present for elevated TSH concentrations (3%–15%, Figure 1D), subclinical hypothyroidism (2%–9%, graph not shown), and overt hypothyroidism (1%–7%, Figure 1F).

Figure 1.

Cross-sectional association of the GRS with TPOAbs, TH markers and clinical outcomes. A, Association of TPOAb concentrations and GRS (GRS distribution in the background). B, Association of TPOAb positivity and GRS. C, Association of TSH concentrations and GRS (GRS distribution in the background). D, Association of elevated TSH concentrations and GRS. E, Association of FT4 concentrations and GRS (GRS distribution in the background). F, Association of over hypothyroidism and GRS.

Subsequently, we evaluated the multivariable-adjusted association between quartiles of the GRS, TPOAb positivity, and hypothyroidism (Table 3). There was a strong, graded, and significant association across risk score quartiles with prevalent TPOAb positivity in all 3 studies. Compared with the lowest GRS quartile (reference), the ORs for TPOAb positivity for those in the highest GRS quartile were 2.10 (95% confidence interval [CI], 1.74–2.55) in the ARIC study and 3.84 (95% CI, 2.50–5.92) and 3.89 (95% CI, 1.73–8.75) for the SHIP-0 and SHIP-T studies. The P values for trend were highly significant in all 3 studies. Similarly, the GRS was significantly associated with 1.2-fold higher odds of elevated TSH per each higher quartile (P < .0001), with similar estimates for subclinical and overt hypothyroidism in the ARIC study, in which sufficient case numbers were available. The OR for overt hypothyroidism was 1.89 (95% CI, 1.24–2.87) for those in the highest GRS quartile compared with that for those in the lowest. Supplemental Table 6 shows associations for the continuous thyroid function measures TPOAbs, TSH, and FT₄ with the GRS: there was a strong, graded, and highly significant association between higher GRS quartiles with TPOAb concentrations in the ARIC and SHIP-0 cohorts and a strong association with higher TSH and lower FT₄ in the ARIC study (P for trend significant for all 3 markers in the large ARIC study). For the smaller SHIP-T cohort, there were no significant associations between the GRS and the evaluated measures, but the effect directions were consistent with those the other studies.

Table 3.

Prevalent Associations Between the GRS and Biomarker-Based Thyroid Outcomes in the ARIC and SHIP Studies

	Q1	Q2	Q3	Q4	Overall (trend)
TPOAb positivity
ARIC
No. case patients/control subjects	186/1826	243/1768	282/1730	353/1658	1064/6982
OR (95% CI)	Ref	1.36 (1.11–1.66)	1.63 (1.33–1.98)	2.10 (1.74–2.55)	1.27 (1.20–1.35)
P value		3.4E–03	1.6E–06	3.1E–14	2.9E–15
SHIP-0
No. case patients/control subjects	28/924	62/887	77/874	98/853	265/3538
OR (95% CI)	Ref	2.35 (1.49–3.72)	3.01 (1.93–4.70)	3.84 (2.50–5.92)	1.47 (1.30–1.65)
P value		2.6E–04	1.2E–06	1.0E–09	2.1E–10
SHIP-T
No. case patients/control subjects	8/214	14/207	17/205	28/194	67/820
OR (95% CI)	Ref	1.79 (0.73–4.40)	2.22 (0.94–5.27)	3.89 (1.73–8.75)	1.55 (1.22–1.97)
P value		2.0E–01	7.0E–02	1.0E–03	3.6E–04
Elevated TSH: ARICa
No. case patients/control subjects	88/1881	127/1837	153/1826	153/1814	521/7358
OR (95% CI)	Ref	1.51 (1.14–2.00)	1.82 (1.39–2.39)	1.82 (1.39–2.39)	1.20 (1.11–1.30)
P value		3.9E–03	1.5E–05	1.6E–05	8.5E–06
Subclinical hypothyroidism: ARICa
No. case patients/control subjects	52/1881	72/1836	93/1825	89/1814	306/7356
OR (95% CI)	Ref	1.46 (1.01–2.09)	1.87 (1.32–2.65)	1.80 (1.27–2.55)	1.21 (1.09–1.34)
P value		4.3E–02	4.0E–04	1.0E–03	4.1E–04
Overt hypothyroidism: ARICa
No. case patients/control subjects	35/1881	55/1836	60/1825	63/1814	213/7356
OR (95% CI)	Ref	1.64 (1.07–2.53)	1.80 (1.18–2.76)	1.89 (1.24–2.87)	1.20 (1.06–1.36)
P value		2.4E–02	6.4E–03	3.1E–03	3.9E–03

^aThe ARIC study only because of sample size.

In the secondary analysis, the association of the GRS with incident TPOAb positivity (OR, 1.21; 95% CI, 1.03–1.43) and subclinical hypothyroidism (OR, 1.14; 95% CI, 1.01–1.28) at visit 5 was of a magnitude comparable to the association with prevalent TPOAb positivity and prevalent subclinical hypothyroidism at visit 2.

Associations between the GRS and thyroid ultrasound findings

Supplemental Table 7 shows a graded and significant association between higher GRS and lower thyroid volume in SHIP-0, which was not observed in SHIP-T. There was a significant inverse association across quartiles of the GRS with lower odds of goiter in SHIP-0 (OR, 0.92; 95% CI, 0.86–0.98), which was also not observed in SHIP-T. Low echogenicity did not show an association with the GRS in either SHIP-0 or SHIP-T.

Discussion

There are 2 principal findings of our study. First, we identified 4 novel genomic regions (1 including the RERE gene and 3 in the extended HLA region) that contain variants associated with TPOAbs at genome-wide significance. We further confirmed associations at 5 known genetic loci. Second, a weighted GRS based on the independent index SNPs at these 9 genetic loci showed strong and graded associations with TPOAbs, TSH, and FT₄ concentrations as well as with the risk of subclinical and overt hypothyroidism and goiter in the general population.

One of the major strengths of GWASs is the ability to map previously unknown risk genes in an unbiased fashion, which can serve as the basis for gaining pathophysiological insights (29). This is true even if the associated variants only explain a modest part of the trait variance as observed here and also if the variants themselves do not significantly improve already well-calibrated clinical prediction models. In our study, we identified 4 novel associated loci. The intronic index SNP rs301799 on chromosome 1p36.23 maps into the RERE gene, which encodes the arginine-glutamic acid dipeptide [RE] repeats protein. The variant identified here has been reported to associate with levels of the RERE transcript (30). A highly correlated genetic variant in this gene (r² of 0.97 in 1000 Genomes CEU data) has previously been reported to be associated with the autoimmune disease vitiligo (31) and thus represents another example of a genetic variant associated with more than 1 autoimmune disease. Individuals with vitiligo show an increased prevalence of AITD (32), and our findings suggest that RERE may be a gene contributing to the shared genetic predisposition (33). The RERE protein is a transcriptional corepressor that is highly expressed in lymphoid cells and is thought to regulate apoptosis (34, 35), which seems to play an important role in the development of AITD, leading ultimately to the destruction of the thyroid gland (36–38).

The other 3 newly genome-wide significant SNPs map into the extended HLA region on chromosome 6, concordant with many already known susceptibility loci for autoimmune diseases including AITD by candidate gene and GWASs (11, 12). The marker rs9277555 maps into HLA-DPB1; genetic variants in HLA-DPB1 have previously been shown to associate with Graves disease (39).

The intergenic marker rs3094228 maps into a region of extended linkage disequilibrium. The closest genes are HCP5 and MICA; other genes include MICB, HLA-B, and HLA-C, all of which are plausible biological candidates. HLA-C was previously demonstrated to have a strong association with Graves disease in Caucasians (40). HLA-B was found to be associated independent of HLA-C with Graves disease and was a risk factor for Hashimoto disease in Han Chinese (40, 41). HCP5 (HLA complex P5) is also a HLA class I gene consistent with a potential role in autoimmunity (42). MICA has been reported to be associated with insulin-dependent diabetes mellitus, celiac disease, and Behçet disease (43–45). The third index SNP on chromosome 6, rs1894407, maps close to TAP2 and HLA-DOB. The membrane-associated protein encoded by TAP2 is a member of the superfamily of ATP-binding cassette transporters. This locus has been implicated in the susceptibility to other autoimmune disease, such as Graves disease, lupus erythematosus, psoriasis, rheumatoid arthritis, insulin-dependent diabetes mellitus, and celiac disease (46–50). HLA-DOB, the other gene mapping into the same linkage disequilibrium block, is thought to also have an important role in autoimmunity and has been associated with various other autoimmune diseases, such as diabetes mellitus type 1, systemic lupus erythematosus, rheumatoid arthritis, and celiac disease (52, 53).

Taken together, the four novel loci highlight the role of the general immune response in different autoimmune diseases. Functional data to illuminate the molecular mechanisms underlying the associations are needed to understand and target disease-specific mechanisms.

A GRS incorporating information from a previous GWASs showed strong associations in the large, independent ARIC study. This observation suggests that winner's curse is unlikely to have had a major impact on results of the published GWASs and that the GRS is likely to perform well in additional independent studies of EA participants. The GRS is based on common TPOAb-associated SNPs and facilitated the identification of a subgroup in the general population with a 2-fold higher prevalence of TPOAb positivity. So far, clinical algorithms, such as the THEA score, have been studied to predict the risk of development of clinical thyroid disease in individuals with first- or second-degree relatives with documented Hashimoto or Graves disease (24). Information on affected relatives was not available in our population-based studies, and we only had information about incident thyroid disease after a long gap between study visits. Future prospective studies are needed to evaluate whether addition of the GRS to clinically validated scores such as the THEA score can improve risk prediction.

The GRS also showed a significant strong and graded association with subclinical hypothyroidism. Subclinical hypothyroidism has increasingly been recognized as an important clinical entity. Besides the fact that it may lead to overt hypothyroidism, subclinical hypothyroidism itself has been associated with hypertension, dyslipidemia, cardiovascular abnormalities, and endothelial dysfunction with consecutive increased cardiovascular morbidity and mortality, reduced fertility, and adverse pregnancy outcomes (2, 51).

Our study has some limitations. Because the ARIC and SHIP studies are population-based studies, there were insufficient case numbers to evaluate subclinical and overt hyperthyroidism, but TPOAbs are clinically considered to be much more a marker of Hashimoto thyroiditis than Graves disease, for which TSH receptor antibodies are more important. Information on goiter, low echogenicity, and thyroid volume was only available for the SHIP cohorts, but outcomes that were available in both ARIC and SHIP showed consistent associations with the GRS despite their locations on different continents with different iodine status of populations. This finding suggests that genetic risk is evident even with variable environmental exposures. Although study heterogeneity and differences in the tests used to assess thyroid function represent limiting factors, the fact that our findings are generally consistent across studies increases confidence related to their validity and generalizability. Because our analyses were restricted to participants of EA descent, our results may not be generalizable to populations with other ancestries.

Because of the unavailability of more closely spaced information on thyroid disease, we could not reliably assess prospective associations of the GRS.

In summary, we identified 4 novel genetic loci and confirmed 5 known loci associated with TPOAbs. A GRS based on index SNPs in these 9 loci showed strong and graded associations not only with TPOAb concentrations and positivity but also with TSH and FT₄ concentrations, as well as with the clinical entities subclinical and overt hypothyroidism and goiter. Our results support the generalizability of the GRS in populations of European ancestry even under different environmental exposures and warrant an evaluation of its predictive ability in combination with clinical patient characteristics in future studies of incident thyroid disorders.