Introduction

Autoimmune thyroid diseases (AITD) encompass a range of thyroid conditions, including hyperthyroidism (Graves’ disease), Hashimoto’s disease (Hashimoto’s thyroiditis), idiopathic myxedema, and primary hypothyroidism [1,2,3]. There is solid evidence that interactions between susceptibility genes and environmental triggers activate the sequence of cellular and humoral immune responses to thyroid antigens that cause AITD [4,5,6]. Several environmental factors, including exposure to excess iodine, selenium deficiency, various infection, certain drugs, and pollutants, have been reported to be associated with AITD [7, 8].

Genetic susceptibility is believed to play a crucial role in this disease etiology [9, 10]. Genetic screening has shown that immune system regulation genes include human leukocyte antigen (HLA) and the cytotoxic T-lymphocyte-associated antigen-4 that are susceptible to AITD [11,12,13]. Genetic-linkage studies have implicated chromosome 8q24 as a strong susceptibility locus for AITD, sequenced all 48 exons of the thyroglobulin (TG) gene, and identified 14 single-nucleotide polymorphisms (SNPs) [14, 15]. As a 660-kDa glycoprotein, TG provides three things: a thyroid hormone precursor, storage of iodine and of inactive thyroid hormones, and acts as a major thyroid autoantigen to map to this region in AITD [15, 16].

The most studied polymorphisms in the TG gene are rs2069550 (E10SNP158), rs853326 (E12SNP), and rs180223 (E10SNP24) in exons 10–12 and rs2076740 (E33SNP) in exon 33 SNP, which were reported to be associated with AITD [17,18,19]. The same study also found that disease risk increased further when combined with the known HLA susceptibility allele, specifically in the DR3 region [18, 20]. Because TG is an important candidate for one of the major autoantigens for AITD, these data are potentially exciting and are likely to lead to further genetic and functional studies. With evidence for either linkage or association has been investigated in many countries and support from mouse models for thyroiditis [21, 22], the roles for the polymorphisms of TG in AITD risk initially seem confused [5, 18, 23]. Therefore, in order to fully understand the role of TG gene polymorphisms in AITD, it is important to review all of the available data at present. A single study may have low statistical power due to small sample size, single ethnicity, and other limitations, but a comprehensive analysis on different studies with different ethnicities will provide strong evidence on the association of polymorphisms in the TG gene and the risk for AITD. Therefore, we performed this meta-analysis of all eligible studies to investigate the association between TG gene polymorphisms and the risk for AITD. This, to our knowledge, is the first meta-analysis addressing this issue.

Results

Characteristics of studies

Appropriate diagnostic criteria and proper genotyping methods were used in all included studies. Ten case–control studies reporting genotypic frequencies both in cases and in controls were included in meta-analysis [5, 18, 19, 23,24,25,26,27,28,29]. Each study character was subdivided into different subgroups. The patient ethnicities included Asians for a majority of studies [5, 19, 24, 26,27,28,29]. There were only two populations with European patients [18, 25] and one population with African patients [23]. The study region latitude was subdivided into four subgroups: “21–30°N” [24], “31–40°N” [5, 19, 23, 26,27,28,29], “41–50°N” [25], and “>50°N” [18]. Age was subdivided into three subgroups: “≦50 years” [19, 23, 25,26,27,28,29], “>50 years” [5], and unknown [18, 24]. All polymorphisms in the control subjects were in Hardy–Weinberg equilibrium (HWE), except one study for rs853326 polymorphism [27], one for rs2076740 polymorphism [19], and one study for rs2069550 polymorphism and rs2076740 polymorphis without data [26]. The results of the studies were given in Fig. 1, and study characteristics were summarized in Table 1.

Fig. 1
figure 1

Flow diagram of the systematic review and meta-analysis literature search results. TG thyroglobulin

Full size image
Table 1 Studies with continuous data on TG polymorphisms in AITD and controls
Full size table

Association between rs2076740 polymorphism and AITD susceptibility

We firstly analyzed the association between rs2076740 polymorphism and the susceptibility to AITD in ten case–control studies that included 3013 cases and 1812 controls. Random-effects model was used to detect the study heterogeneity. The analysis showed that rs2076740 polymorphism was significantly associated with AITD in CT vs. TT codominant model (odds ratio (OR) = 0.757, P = 0.005) (Fig. 2a). The estimated OR1, OR2, and OR3 were 0.606, 0.757, and 0.950, respectively. These estimates suggested a codominant genetic model. The pooled OR1 was 0.606 (P = 0.058). There was no evidence of publication bias detected by Egger’s test (Egger P = 0.104). The sensitivity analysis by removal of two study [23, 26] changed the results in CC+CT vs. TT dominant models. The recalculated OR was 0.683 (P = 0.016). These results indicated that the CC genotype might be a significant AITD-protective factor compared to the TT genotype (Table 2).

Fig. 2
figure 2

Meta-analysis with a fixed-effects model for the association between rs2076740 polymorphism and AITD risk in the codominant model

Full size image
Table 2 Association between rs2076740, rs853326, rs180223, and rs2069550 TG polymorphisms and AITD risk
Full size table

The subanalysis showed that AITD risk was significantly reduced in the “31–40°N” group (OR = 0.682, P = 0.012) in the CT vs. TT codominant model, which indicates CT genotype has a protective effect in the “31–40°N” subgroup (Fig. 2b). The results are summarized in Table 3.

Table 3 Association between rs2076740 polymorphism and AITD: stratification according to study characteristics
Full size table

Association between rs180223 polymorphism and AITD susceptibility

The association between rs180223 polymorphism and the risk of AITD was analyzed in five independent studies with 1896 cases and 1025 controls. Random-effects model was used in the dominant model, codominant model, and allele model due to the presence of heterogeneity, and fixed-effects model was used in the recessive model and homozygous model. The analysis showed that rs180223 polymorphism was not associated with AITD in any of the studied models. The estimated OR1, OR2, and OR3 were 1.156, 1.442, and 0.884, respectively. These estimates suggest a codominant model, and the pooled OR1 = 1.156 (P = 0.278), OR2 = 1.442 (P = 0.298), and OR3 = 0.884 (P = 0.638). The TT genotype was a significant AITD risk factor. There was no heterogeneity in TT vs. GG (I2 = 0%). The sensitivity analysis by removal of each individual study did not meaningfully change the results in the codominant model. There was no evidence of publication bias detected by the Egger’s test in TT vs. GG (Egger P = 0.949) (Table 2).

Association between rs2069550 polymorphism and AITD susceptibility

The association between rs2069550 polymorphism and the risk of AITD was analyzed in six independent studies with 2349 cases and 1433 controls. Fixed-effects model was used in all models due to the presence of heterogeneity. The analysis showed that rs2069550 polymorphism was not associated with AITD in any of the studied models. The estimated OR1, OR2, and OR3 were 1.108, 0.918, and 1.169, respectively. These estimates suggested a recessive genetic model. The pooled OR was 1.152 (P = 0.22). The sensitivity analysis by removal of one study [18] changed the results in T vs. C allele model. The recalculated OR was 0.844 (P = 0.024) (Fig. 3). These results indicated that the T genotype might be a significant AITD-protective factor compared to the C genotype (Table 2).

Fig. 3
figure 3

Meta-analysis with a fixed-effects model for the association between rs2069550 polymorphism and AITD risk in the allele model

Full size image

The subanalysis showed that people in 31–40°N latitudes significantly reduced AITD risk in both allele model C vs. T (OR = 0.823, P = 0.012) and TT + TC vs. CC dominant model (OR = 0.773, P = 0.009) (Fig. 4a, b). The Aisan people with allele model C vs. T and TT+TC vs. CC dominant model also tended to protect against AITD (OR = 0.799, P = 0.012; OR = 0.773, P = 0.009) (Fig. 4c, d) (Table 4).

Fig. 4
figure 4

Meta-analysis with a fixed-effects model for the association between rs2069550 polymorphism and AITD risk in the allele and dominant model

Full size image
Table 4 Association between rs2069550 polymorphism and AITD: stratification according to study characteristics
Full size table

Association between rs853326 polymorphism and AITD susceptibility

The association between rs853326 polymorphism and the risk of AITD was analyzed in seven independent studies with 2379 cases and 1255 controls. Random-effects model was used in the homozygous model, dominant model, codominant model (OR1 and OR2), and allele model due to the presence of heterogeneity, and fixed-effects model was used in the recessive model and codominant model (OR3). The estimated OR1, OR2, and OR3 were 0.854, 0.856, and 1.187, respectively. These estimates suggested a dominant genetic model. The pooled OR was 1.020 (P = 0.94). The sensitivity analysis by removal of one study [19] changed the results in AA + AG vs. GG dominant model. The recalculated OR was 1.205 (P = 0.027) (Fig. 5a). These results indicated that the AA genotype might be a significant AITD risk factor compared to the GG genotype (Table 2).

Fig. 5
figure 5

Meta-analysis with a fixed-effects model for the association between rs853326 polymorphism and AITD risk in the dominant model

Full size image

Discussion

The 8q24 region to which the TG gene maps was shown to be strongly linked with AITD [30]. Ban et al. [31] demonstrated that an exon 10–12 SNP cluster and an exon 33 SNP were significantly associated with AITD. However, Collins et al. [18] stated that the SNPs in exons 10, 12, and 33 do not have a causal role for AITD in the United Kingdom in a study of 1214 Caucasian patients in the United Kingdom with AITD. Thus, the association between TG gene polymorphisms and AITD is still controversy. This meta-analysis firstly evaluate the genetic associations between rs2076740, rs180223, rs2069550, and rs853326 polymorphisms and AITD (GD and HT).

Consistent with Hsiao et al.’s study [28], our analysis clearly revealed the rs2076740 was significantly associated with AITD in the CT vs. TT codominant model. This finding suggests that the homozygote CC genotype might be a significant AITD risk factor. In our study, rs180223, rs2069550, and rs853326 polymorphisms were not implicated in AITD. However, the sensitivity analysis indicated instability of results and showed significant associations between the dominant model of rs2076740, the allele contrast model of rs2069550, and the homozygous model of rs853326 polymorphisms with AITD risk. These results suggest that the TG gene is significantly involved in AITD. Additionally, recent genetic association studies have provided evidence of the involvement of AITD-associated genes and Stefan et al. [32] identified a −1623A/G SNP (rs180195) in the promoter region of the TG gene that modified a binding site for interferon regulatory factor-1 (IRF-1) or ETS transcriptions factor-1 by combinating the G allele in TG at the 8q24.22 region predisposed to AITD. Therefore, AITD is generally considered as the result of interactions among multiple genes.

Considering that AITD is widely accepted as a complex trait that develops in genetically susceptible persons exposed to environmental risk factors [4, 7, 33], we stratified our studies according to study characteristics. Study insufficiency is considered a major limitation when evaluating and comparing all interactions between TG polymorphisms and study characteristics. We found that the associations between rs2069550 polymorphism and AITD risk was successively dependent on ethnicity. Additionally, we found that 31–40°N latitude significantly affected the association between rs2076740 and rs2069550 polymorphisms and AITD risk; it may conform to Martinez’s hypothesis [34] that some relationships between genotype and disease will only be observed in conditions of “high” exposure to an environmental factor of interest and others may only be observed in conditions of “low” exposure. Therefore, genetically predisposed individuals with polymorphisms in genes important for TG metabolism, catabolism, or function has an increased likelihood of developing autoimmune diseases [5, 16, 19].

There are several caveats in this meta-analysis that should be discussed. Most of the group-level data showed large between-study heterogeneity. This could be due to bias, chance, or genuine diversity of genetic effects. Based on Egger’s test, we detected a small publication bias. The bias might be caused by unpublished data because negative studies were less likely to be published in journals and were not available in a computerized database, and only studies indexed by the selected databases were included. This results in a potential overestimation of effect sizes. In addition, the smaller published studies did not show different results compared with the larger ones. However, there was a suggestion that the first studies may have yielded somewhat stronger effects. This is consistent with a “winner’s curse phenomenon” in which early data show exaggerated effects [35]. Thus, the group-level-derived estimates may be modestly inflated, as suggested also by the trend for relatively smaller effects sizes for the allele of rs2076740, rs180223, rs2069550, and rs853326 in the individual-level data. The latter may provide more reliable estimators and more options for deciphering the relative contribution of each polymorphism, but they are also not necessarily devoid from potential biases. Nevertheless, genotyping error for SNPs should be low at experienced facilities. Not all cases of genotype frequencies were consistent with HWE in the individual-level database, which may be attributed to the mutation of TG. Besides biases, this could be attributed to differences in terms of disease phenotype (e.g. presence of type 1 diabetes and/or of other autoimmune diseases) among AITD cases [25]. AITD is rare in men to allow evaluation of gender differences [36]. Moreover, we detected moderate heterogeneity, which is caused by several factors such as differences in ethnicities. Therefore, the results could be influenced by factors such as random error.

In conclusion, despite these caveats, our collaborative analysis shows consistent associations between AITD with TG. This association crosses ethnic barriers and latitude gradients, and we can make a reasonable estimate of the important role of the TG locus in determining the risk of AITD. However, we still cannot identify a single etiological polymorphism because the current study could not evaluate all interactions between-study characteristics and TG polymorphisms due to insufficient information from the primary publications.

Methods and materials

Identification of eligible studies

The review process followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [37]. We performed a literature search in PubMed, Cochrane library, Web of Science and Embase to identify articles that examined the association between rs2076740, rs853326, rs180223, and rs2069550 TG polymorphisms and AITD (Graves’ disease and Hashimoto’s thyroiditis) (updated to May 2018). Combinations of keywords, such as “thyroglobulin polymorphisms” and “Autoimmune thyroid disease” were entered as medical subject heading (MeSH) and text words. The reference lists of the articles retrieved were also reviewed to identify publications on the same topic. Two independent reviewers performed searching in duplicate.

Inclusion and exclusion criteria

Studies in this meta-analysis must meet the following inclusion criteria: (1) evaluation of the association between rs2076740, rs853326, rs180223, and rs2069550 TG polymorphisms and the AITD risk; (2) case–control study; (3) studies focusing on human being; (4) detailed genotype data could be acquired to calculate the ORs and 95% confidence intervals (CIs); Exclusion criteria: (1) duplication of previous publications; (2) comment, review and editorial; (3) family-based studies of pedigrees; (4) study with no detailed genotype data. Study selection was achieved by two investigators independently, according to the inclusion and exclusion criteria by screening the title, abstract and full-text. Any dispute was solved by discussion.

Data extraction

The data of the eligible studies were extracted in duplicate by two investigators independently (Zhang and Chen). The following contents were collected: name of first author, year of publication, the characteristics of cases and controls, country of origin, the detective sample, ethnicity, genotyping methods, HWE, number of cases and controls in case and control for rs2076740, rs853326, rs180223 and rs2069550 genotypes, respectively. Different ethnicity descents were classified as Caucasian and Asian. Two authors checked the extracted data and reached to consensus on all the data. If a dissent existed, they would recheck the original data of the included studies and have a discussion to reach consensus. If the dissent still existed, the third investigators would be involved to adjudicate the disagreements (Wang).

Statistics analysis

The data from studies were combined to produce a summary OR and represented as an estimated value and 95% confidence intervals (CIs) on a forest plot. The methodology for meta-analysis of molecular studies was described by Thakkinstian et al. [38]. The OR1, OR2, and OR3 values were calculated for the following genotypes: (i) CC vs. TT (OR1), CT vs. TT (OR2), and CC vs. CT (OR3) for the rs2076740 polymorphism; (ii) TT vs. GG (OR1), TG vs. GG (OR2) and TT vs. TG (OR3) for rs180223 polymorphism; (iii) TT vs. CC (OR1), TC vs. CC (OR2), and TT vs. TC (OR3) for rs2069550 polymorphism; and (iv) AA vs. GG (OR1), AG vs. GG (OR2), and AA vs. AG (OR3) for the rs853326 polymorphism. The pairwise differences were used to indicate the most appropriate genetic model as follows: if OR1 = OR3≠1 and OR2 = 1, then a recessive model was suggested; if OR1 = OR2≠1 and OR3 = 1, then a dominant model was suggested; if OR2 = 1/OR3≠1 and OR1 = 1, then a complete overdominant model (homozygous) was suggested; if OR1 > OR2 > 1 and OR1 > OR3 > 1 (or OR1 < OR2 < 1 and OR1 < OR3 < 1), then a codominant model was suggested [38]. The data heterogeneity was evaluated using the Q-statistic [39]. When the significant Q-statistic indicated heterogeneity across studies, then a random-effects model was used. I2 values of 25%, 50% and 75% were defined as low, moderate, and high estimates, respectively. A symmetric plot and the P value of Egger’s test <0.05 was considered a significant publication bias [40]. The subgroup analysis was planned when sufficient information was reported in at least two studies in each subgroup. The stability of the summary risk estimate was evaluated using a sensitivity analysis in which each study was individually removed and the OR was recalculated. All statistical analyses were performed with the Stata 12.0 software (StataCorp, College Station, TX, USA). A two-tailed P < 0.05 was considered as significant except for specified conditions, where a certain P value was declared.