Thyroid carcinoma is the most common endocrine malignancy, and its incidence in industrialized countries has been rising over the past few decades: at present, its incidence is 4.9 and 14.1 per 100,000 in the United States for males and females of European ancestry, respectively (see URLs section in Methods). It has been determined that the risk of thyroid cancer has a greater genetic component than the risk of any other cancer, and the effect has been shown to extend beyond the nuclear family^1–3. The risk has been reported to be highest for first-degree male relatives of male probands but lowest for first-degree female relatives of female probands⁴. Not much is known about variants in the sequence of the genome that affect the risk of thyroid cancer. However, recently variants at 1p12, 8q24 and in the pre-miR146a at 5q33 have been implicated in the disease^5–8.

Thyroid cancer is classified histologically into four main groups: papillary (PTC), follicular (FTC), medullary (MTC) and undifferentiated or anaplastic thyroid carcinomas. Most of all thyroid tumors are PTC (80–85%) or FTC (10–15%)^9,10. Among established risk factors for FTC is deficiency in iodine intake, and for PTC the risk factors include ionizing radiation, nodular disease of the thyroid as well as family history⁹. Medullary thyroid cancer is a part of the multiple endocrine neoplasia type 2 syndrome (MEN2) and accounts for about 1–3% of all thyroid cancer cases. Its pathogenesis can mainly be explained by mutations in the RET oncogene¹¹. Anaplastic thyroid cancer accounts for between 1% and 5% of all carcinomas of the thyroid. This subphenotype is one of the most aggressive of all human malignancies, but little is known about its pathogenesis¹².

In Iceland, the annual incidence of thyroid cancer is similar to that in the United States, at 4.6 and 12.1 per 100,000 for males and females, respectively, according to the Icelandic Cancer Registry (see URLs section below). In order to search for sequence variants conferring risk of thyroid cancer, we conducted a genome-wide association study (GWAS) with 192 histopathologically confirmed Icelandic thyroid cancer cases and 37,196 controls genotyped using the Illumina HumanHap300 and HumanCNV370-duo Bead Chip genotyping platform (Supplementary Methods online). Furthermore, we used a method in which known genotypes of relatives are used to provide information on individuals with thyroid cancer who were not genotyped (in silico genotyping), in order to add genotypes that represent, on average per SNP, an additional 186 individuals with thyroid cancer¹³. After removal of SNPs that failed quality checks, a total of 304,083 SNPs were tested for association. We calculated the allelic odds ratio (OR) for each SNP assuming the multiplicative model, and computed a standard likelihood ratio χ² statistic for the purpose of testing. The results were adjusted for familial relatedness between individuals and for potential population stratification using the method of genomic control¹⁴; the χ² statistics were divided by an estimated inflation factor of 1.09. Quantile-quantile plots of the χ² statistics before and after adjustment are shown in Supplementary Figure 1 online.

The nine strongest signals (P < 1 × 10⁻⁶) are shown in Supplementary Table 1 online; of these, seven were located in the same linkage disequilibrium (LD) region as the FOXE1 (forkhead factor E1) gene on 9q22.33 (Fig. 1). In an attempt to confirm these results we proceeded to genotype the nine SNPs in additional 241 Icelandic thyroid cancer cases using Centaurus¹⁵ single-track assay genotyping. Combining these results and the results from the GWAS, we observed that the strongest association signals were for allele A of rs965513 (rs965513[A]) and allele A of rs10759944 (rs10759944[A]), with an OR of 1.77 for both variants (P = 6.8 × 10⁻²⁰ and P = 1.7 × 10⁻¹⁹ for rs965513 and rs10759944, respectively; Table 1 and Supplementary Table 1). These two SNPs are nearly perfect surrogates of each other (r² = 1 in the Utah CEPH (CEU) HapMap samples and r² = 0.998 in the Icelandic samples), and because the effects of the variants cannot be distinguished from each other, we elected to focus on rs965513[A] in subsequent investigations. When we controlled for rs965513[A] in a multivariate analysis, none of the remaining SNPs on 9q22.33 (ranking 3 to 7 in Supplementary Table 1) is significant. In the combined analysis of the Icelandic data, the eighth-ranking association signal was observed for allele T of rs944289 (rs944289[T]) located on 14q13.3. This signal reached a genome-wide significance, with an OR of 1.37 and P = 2.5 × 10⁻⁸ (Table 1 and Supplementary Table 1). The ninth SNP, rs622450 on 1p36.13, failed to replicate in the Icelandic samples and was not studied further (Supplementary Table 1).

Open in a separate window

Figure 1

A schematic view of the association results and LD structure in a region on chromosome 9q22.33. (a) Single-marker (blue diamonds) association results for SNPs from the Illumina Hap300/370 chip. The results are based on analysis of 192 and 37,217 Icelandic thyroid cancer cases and controls, respectively, and shown are P values corrected for relatedness. (b) Pairwise correlation coefficient (r²) from the CEU HapMap population and the relative location of genes in the region, based on the UCSC Genome Browser, build 36. The locations of rs965513 and rs10759944 are not resolved in this image.

We next tested the association of the two genome-wide significant variants, rs965513 on 9q22.33 and rs944289 on 14q13.3, with thyroid cancer in two case-control groups of European descent, with populations from Columbus, Ohio, United States (342 cases and 384 controls) and Spain (90 cases and 1,343 controls). The results for both SNPs were replicated in both study groups (Table 1). A test of heterogeneity in the ORs between the three study populations showed no significant difference (P = 0.58 for rs965513 and P = 0.35 for rs944289). Combination of the results from Iceland, Columbus and Spain gave an estimated OR of 1.75 for rs965513[A] (P = 1.7 × 10⁻²⁷), whereas rs944289[T] gave an OR of 1.37 (P = 2.0 × 10⁻⁹) (Table 1).

In order to investigate the mode of inheritance, we computed the genotype-specific ORs and found that the multiplicative model provided an adequate fit for both variants (Supplementary Table 2 online). Approximately 11% and 32% of individuals in the general population are homozygous carriers of rs965513[A] or rs944289[T], respectively, and 3.7% are homozygous for both variants. Homozygous carriers of rs965513[A] or rs944289[T] are estimated to have 3.1- and 1.9-fold greater risk, respectively, of developing the disease than noncarriers. Doubly homozygous individuals have an estimated 5.7-fold greater risk. Furthermore, we observed that the frequency of rs965513[A] was higher among subjects diagnosed at a younger age in all three populations. With the data combined, it is estimated that, for each allele carried, age at diagnosis is reduced by 2.42 y (P = 0.0014; Supplementary Table 3 online). No significant age effect was observed for rs944289 and no significant difference was seen, for either variant, between the observed effects on males and females (P = 0.19 and P = 0.56 for rs965513[A] and rs944289[T], respectively; Supplementary Table 3).

We analyzed the effect of rs965513 and rs944289 in the four main histological classes of thyroid cancer. Most of the Spanish and Icelandic sample collections consist of PTC (~85%) and FTC (~12%) and all of the cases from Columbus were PTC. For rs965513[A] on 9q22.33, the observed OR for PTC in the combined analysis of the three populations was 1.80 (P = 4.7 × 10⁻²³) and for FTC the OR was 1.55, calculated using the Icelandic and Spanish samples only (P = 0.016) (Supplementary Table 4 online). Similarly, for rs944289[T] on 14q13.3, the OR for the PTC in the combined analysis of all three study populations was 1.32 (P = 2.0 × 10⁻⁶) and for FTC the OR was 1.63, calculated from the Icelandic and Spanish samples only (P = 0.0071) (Supplementary Table 4). This demonstrates that both variants affect the risk of the two main histological types of thyroid cancer. The numbers of other histological thyroid cancer types were too limited to draw meaningful conclusions.

The SNP rs965513 resides on 9q22.33 within a LD region where the following genes have been localized: XPA, FOXE1, C9orf156 and HEMGN (Fig. 1). The closest gene is FOXE1, located about 57 kb telomeric to rs965513. FOXE1 is important for both pituitary- and thyroid-gland formation^16,17 and is at the center of a regulatory network of transcription factors and cofactors that initiate thyroid differentiation at the embryonic stage¹⁸. Furthermore, mutations of the FOXE1 gene cause human syndromes that are associated with thyroid agenesis, among other phenotypes^17,19. FOXE1 is also necessary for the maintenance of the differentiated state of the thyroid, as it is involved in regulating the transcription of thyroid-specific genes, such as the TG (thyroglobulin) and TPO (thyroperoxidase) genes. Regulated expression of both of these genes is pivotal for the synthesis of the thyroid hormones triiodothyronine (T₃) and thyroxine (T₄), as Tg is the precursor of the T₃ and T₄, and their synthesis is catalyzed by TPO. Central to the thyroid hormone synthesis and secretion control is the thyroid stimulating hormone (TSH), which acts as principal regulator.

The 14q13.3 SNP rs944289 is in a 249-kb LD-region with no RefSeq genes. The closest genes are: BRMS1L, MBIP, SFTA3 and NKX2-1(TTF1) (Supplementary Fig. 2 online). Although several of these genes have been implicated in cancers at various sites, NKX2-1 is probably the best candidate as a source of the association signal, because it has a prominent role in the development of the thyroid¹⁸ and its expression is altered in thyroid tumors²⁰.

Given the involvement of FOXE1 and NKX2-1 in the biology of the thyroid gland, we assessed the effect of rs965513[A] and rs944289[T] on circulating levels in serum of TSH (N = 12,035), free T₄ (N = 7,108) and free T₃ (N = 3,593). The data used came from a series of measurements collected over a period of 11 y (from 1997 to 2008) from Icelanders not known to have thyroid cancer (Supplementary Table 5 online). Both variants were associated with (P = 2.90 × 10⁻¹⁴ and P = 0.03 for rs965513[A] and rs944289[T], respectively) a decrease in serum concentrations of TSH, by 5.9% and 1.7% per copy of rs965513[A] and rs944289[T], respectively (Table 2). rs965513[A] on 9q22.33 was also associated with serum concentrations of T₃ and T₄, but in opposite directions; each copy of rs965513[A] was associated with an increase in T₃ levels by 1.2% and a decrease in T₄ levels by 1.2% (P = 3.00 × 10⁻³ and 6.10 × 10⁻⁵ for T₃ and T₄, respectively) (Table 2). For rs944289[T] on 14q13.3, no significant effect on either T₃ or T₄ was observed (Table 2). These data demonstrate that at least the 9q22.33 variant affects some aspects of the endocrine function of the thyroid.

Taken together, the effect of the 9q22.33 SNP (rs965513) on thyroid and thyroid-related hormones, the proximity of rs965513 to FOXE1, and the controlling effect of FOXE1 on thyroid-specific genes strongly suggest that the association between thyroid cancer and rs965513 is mediated through processes involving FOXE1. Furthermore, the expression of FOXE1 has been shown to be abnormal in thyroid tumors²¹. The explanation of the 14q13.3 association signal seems to be similar and mediated through NKX2-1, although the data are less compelling. Compared to the common susceptibility variants recently discovered for complex diseases, the relative risks associated with the variants reported here are quite high. Together, they have a population attributable risk (PAR) of 57% for populations of European descent. Furthermore, we have demonstrated that these variants influence crucial players in the thyroid hormonal pathway. These variants are therefore likely to be among the most important determinants of genetic susceptibility to thyroid cancer.