The contribution of genetics to the risk of thyroid cancer is greater than to any other cancer, and the effect extends beyond the nuclear family1–4
. Thyroid cancer is classified into four main histology groups: papillary (PTC), follicular (FTC), medullary (MTC), and undifferentiated or anaplastic thyroid carcinomas. The great majority of malignant thyroid tumors are nonmedullary, either PTC (80–85%) or FTC (10–15%)5,6
Among sequence variants that have been implicated in the etiology of PTC are variants at 1p12, 8q24, and the pre-miR146a at 5q33 (refs. 7–10
). Furthermore, in results from a genome-wide association study (GWAS) on thyroid cancer, two common variants, located on 9q22.33 and 14q13.3, were shown to associate with both PTC and FTC11
. The association with variants on 9q22.33 has been replicated both in sporadic12
thyroid cancers. The two variants have also been associated with low serum concentrations of thyroid-stimulating hormone (TSH), and the 9q22.33 variant associates with low concentration of free thyroxin (FT4
) and high concentration of free triiodothyronine (FT3
. Here we focused our search for additional thyroid cancer risk variants on SNPs with P
≤ 5 × 10−8
in a GWAS on circulating TSH levels.
The GWAS on TSH levels included 27,758 individuals not known to have thyroid cancer, on the basis of information from the nationwide Icelandic Cancer Registry (, Supplementary Table 1
and Supplementary Fig. 1
). Genotyping was done using the Illumina SNP chip platform. We also made use of results from genome-wide sequencing of 457 Icelanders to an average depth of over 10× (see and the Supplementary Note
), resulting in the identification of some 16 million SNPs. Using imputation assisted by long-range haplotype phasing14,15
, we inferred the genotypes of these SNPs in 41,675 Icelanders who had been genotyped using the Illumina SNP chip platform, including the 27,758 with TSH measurements.
Study design and results. The number of individuals with thyroid cancer (affected) or unaffected individuals imputed or successfully genotyped is shown for each study segment.
On the basis of an association analysis of this data set, we found 22 variants that associate with serum levels of TSH at a significance threshold of P
< 5 × 10−8
(Supplementary Table 2
). For one of these regions, 9q22.33, a variant (rs965513) associating with both thyroid cancer and TSH levels has been reported11
and will not be discussed any further here. Of the remaining 21 TSH-associated loci, three associate with levels of TSH—1p36 (ref. 16
), 5q14.1 (refs. 17,18
) and 6q27 (ref. 17
)—but none has been reported to associate with thyroid cancer.
By inspecting our GWAS data set on Icelandic nonmedullary thyroid cancer, generated with both chip and imputed genotypes (see Online Methods for a detailed description), 5 of the 21 TSH-associated SNPs were found to associate with thyroid cancer with nominal significance (P
< 0.05, Supplementary Table 2
). We also searched for stronger or additional thyroid cancer association signals by examining the thyroid cancer GWAS results for SNPs located within a 1.5-Mb region centered on each of the 21 original TSH SNPs. On the basis of this analysis, we found a SNP (rs966423) on 2q35 with a more significant association with thyroid cancer than the initial TSH-associated SNP (rs737308) at this locus (P
= 0.0010 and 0.31 for rs966423 and rs737308, respectively). The pairwise correlation between the two SNPs on 2q35 was low (r2
= 0.003, D′
= 0.08, according to data from 2,349 Icelanders), and therefore, they are likely to represent different association signals. rs966423 also associated with serum levels of TSH, but the effect was smaller and the significance less than for rs737308 (Supplementary Table 2
). We found no additional variants associated with thyroid cancer at the remaining TSH loci.
We genotyped the 21 TSH-associated SNPs and the one suggestive thyroid cancer SNP (rs966423) on 2q35 in 561 individuals with thyroid cancers and 3,190 controls from the general Icelandic population, using the Centaurus19
single-track genotyping assay. Four of the TSH-associated SNPs are present on the Illumina chips, and for those, we made use of chip-derived genotypes from 39,864 individuals (see Online Methods for a detailed description). Analysis of directly genotyped individuals (using either the single-track assay or the Illumina chip platform) resulted in a nominally significant (P
< 0.05) association signal for thyroid cancer and five SNPs located on 1p31.3, 1p36.13, 2q35, 8p12 and 14q13.3 (Supplementary Table 2
). At 2q35, only the variant rs966423 was significantly associated with thyroid cancer, whereas rs737308, the SNP on 2q35 with a stronger association with TSH levels, was not (P
= 3.8 × 10−4
and 0.26 for rs966423 and rs737308, respectively).
Four of the five variants were genotyped and tested for association with thyroid cancer in three case-control groups of European descent, with populations from the USA (Ohio), The Netherlands and Spain. The fifth variant, rs10799824 on 1p36.13, was genotyped only in the Dutch and Spanish samples, but the association was not replicated, and this SNP was not studied further. The results for the variant on 1p31.3 (rs334725) did replicate in two out of the three case-control groups, resulting in a combined allelic odds ratio (OR) of 1.31 (P
= 6.6 × 10−3
; Supplementary Table 3
). Whether this variant truly confers risk of thyroid cancer remains to be shown. Combining the results from Iceland and the follow-up groups gave OR estimates between 1.34 and 2.09 for the remaining three variants located on 2q35, 8p12 and 14q13.3 (P
< 3 × 10−9
Association results for three variants on 2q23, 8p12 and 14q13.3 and thyroid cancer in Iceland, the Netherlands, spain and the United statesa
Of the three genome-wide significant thyroid cancer variants reported here, the strongest association was observed for allele T of rs116909374 (rs116909374[T]) on 14q13.3. This SNP and the SNP rs944289, which has been reported to associate with thyroid cancer11
, are located within two distinct but neighboring LD regions (Supplementary Fig. 2
). The correlation between them is very low (r2
= 0.005, D′
= 0.35, according to data from 3,693 Icelanders), and the association with thyroid cancer for each SNP remains significant after adjusting for the other (Supplementary Table 4
). The association effect for TSH levels is substantially stronger for the current SNP than for the previously reported one (effect = 0.141 s.d. and P
= 1.1 × 10−16
for rs116909374[T] compared to an effect = 0.022 s.d. and P
= 0.001 for rs944289[T]). This suggests that the14q13.3 locus contains more than one variant causing a predisposition to thyroid cancer or, alternatively, that a unique variant capturing the effect of rs116909374 and rs944289 remains to be discovered. On 14q13.3, the gene closest to rs116909374 is MBIP
(a regulatory protein), but another nearby candidate gene, which must be considered because of its prominent role in the development of the thyroid20
, is the thyroid transcription factor NKX2-1
. The variant rs966423, at 2q35, is located in the DIRC3
gene and has not previously been associated with thyroid cancer or serum levels of TSH.
The SNP rs2439302, on 8p12, is located within the first intron of the gene NRG1
. The NRG1
gene encodes neuregulin 1, a signaling protein that mediates cell-cell interactions and plays an important role in the development of the nervous system, heart, breast and other organs. Germline sequence variation at the NRG1
locus has been associated with schizophrenia21
and with Hirschsprung’s disease in individuals of Asian descent22
. The thyroid cancer variant reported here (rs2439302[G]) and one of the two originally reported variants (rs7835688[C]) for the Hirschsprung’s disease are positively correlated on the basis of HapMap data from both Asian (CHB+JPT; r2
= 0.95) and European-descended (CEU; r2
= 0.94) individuals. Hence, it is likely that the same sequence variant in NRG1
affects circulating TSH levels and risks of Hirschsprung’s disease and thyroid cancer.
To determine whether any of the three thyroid cancer variants associate with expression levels of the genes in which they are either located or close by, we examined our previously described microarray expression data set23
from whole blood of 966 population-based Icelandic controls. No significant association was detected between the variants on 2q35 and 14q13.3, and expression levels of genes located within a 1-Mb region, centered on the thyroid cancer risk variants. However, a highly significant association was observed between rs2439302, on 8p12, and expression measured by several different probes in the NRG1
gene, the only Reference Sequence (RefSeq) database gene within a 1-Mb region centered on rs2439302. Carriers of the allele conferring risk of thyroid cancer (rs2439302[G]) have a lower relative expression of NRG1
. For the probe with the most significant association (P
= 9.1 × 10−91
), the relative expression of NRG1
was decreased by 40% per each allele G of rs2439302. Hence, the relative expression of NRG1
is 64% lower in homozygous carriers of the thyroid cancer risk allele than in homozygous noncarriers (). The strong association results seen for rs2439302 and thyroid cancer on one hand, and the expression levels of NRG1
on the other hand, strongly suggest that NRG1
plays a role in the etiology of nonmedullary thyroid cancer.
Figure 2 Correlation between relative expression of NRG1 in blood (y axis) and genotypes (x axis) of rs2439302 on 8p12. The expression of NRG1 is shown as 10MLR, where MLR is the mean log expression ratio per individual. In the plot, the bottoms and tops of the (more ...)
Using only the Icelandic study group, no significant effect was detected for age at diagnosis for any of the variants reported here. In terms of effect of the new variants on the histological subclasses of thyroid cancer, the number of FTC samples were, in general, too few to draw meaningful conclusions.
For the SNPs rs966423, rs2439302 and rs116909374, the sibling recurrence risk ratio was estimated to be 1.021, 1.023 and 1.033, respectively, on the basis of combined effect and combined frequencies from all four study groups. In order to summarize the overall effect of these three variants and the two previously reported variants on 9q22.33 and 14q12.3, we combined the effect of all five variants known today to confer risk of thyroid cancer. We used the combined risk estimates for the four study population used in the present study, assuming a multiplicative model for the five risk variants. On the basis of this analysis, the estimated risk of the disease was over 2.3 times greater for the top 5% of the risk distribution, than for the general population. Similarly, the risk of the disease was over three times greater for the top 1% of the risk distribution. We note that the estimates provided here are based solely on populations used in this study and will have to be updated as new variants are discovered.
For the 22 variants associating with TSH levels at a significance threshold of P
< 5 × 10−8
, the fraction of variance of TSH explained was 4.3%. We also checked their association with risk of goiter and with levels of FT4
. Of the 22 TSH loci, 9 associated with risk of goiter with nominal significance (P
< 0.05). Two of these loci, 1p36.13 and 15q21.1, were recently reported in a GWAS on goiter risk24
. Two of the TSH variants had a nominally significant association with levels of FT3
(rs965513 on 9q22.33 and rs61938844 on 12q23.1; P
< 0.05), and 12 of the 22 TSH variants had a nominally significant association with levels of FT4
Taken together, these results demonstrate the unresolved complexity of DNA polymorphism and their effects on biological functions such as levels of thyroid-related hormones and risk of both benign and malignant thyroid diseases. The variants associating with serum TSH concentrations can be divided into two main groups; those that do and those that do not confer risk of thyroid cancer (Supplementary Fig. 3
). Notably, for all the variants that confer risk of thyroid cancer, the at-risk allele for thyroid cancer associated with lower serum levels of TSH. This was true for the three variants reported here as well as for the two thyroid cancer-TSH variants previously published11
. Furthermore, the variants on 2q35 and 14q13.3 reported above also associated with increased levels of FT4.
Hence, it appears that for carriers of these five variants the primary disorder in nonmedullary thyroid cancer is an endocrine one, characterized by a lower concentration of TSH. The consequence of the low concentration of TSH may be less differentiation of the thyroid epithelium, leading to a predisposition to malignant transformation. There is, however, more to the story of thyroid cancer because not all sequence variants associating with low TSH levels associate with the risk of the disease.