Genetic–epigenetic dysregulation of thymic TSH receptor gene expression triggers thyroid autoimmunity

Significance

Graves disease (GD) is an autoimmune disease caused by interactions between genetic, epigenetic, and environmental factors. The thyrotropin receptor (TSHR) is the major autoantigen in GD and is a key GD susceptibility gene. SNPs in intron 1 of the TSHR are associated with GD, but the causative variant and the mechanisms are unknown. By mapping epigenetic modifications induced by IFNα, a viral-induced cytokine triggering GD, we pinpointed the causative variant in intron 1 of the TSHR. We demonstrate that the disease-associated variant interacts epigenetically with a transcriptional repressor, promyelocytic leukemia zinc finger protein, and reduces thymic TSHR expression, leading to escape from tolerance and autoimmunity to the TSHR. These genetic–epigenetic interactions leading to decreased thymic self-antigen expression reveal a universal mechanism in autoimmunity.

Abstract

Graves disease (GD) is an autoimmune condition caused by interacting genetic and environmental factors. Genetic studies have mapped several single-nucleotide polymorphisms (SNPs) that are strongly associated with GD, but the mechanisms by which they trigger disease are unknown. We hypothesized that epigenetic modifications induced by microenvironmental influences of cytokines can reveal the functionality of GD-associated SNPs. We analyzed genome-wide histone H3 lysine 4 methylation and gene expression in thyroid cells induced by IFNα, a key cytokine secreted during viral infections, and overlapped them with known GD-associated SNPs. We mapped an open chromatin region overlapping two adjacent GD-associated SNPs (rs12101255 and rs12101261) in intron 1 of the thyroid stimulating hormone receptor (TSHR) gene. We then demonstrated that this region functions as a regulatory element through binding of the transcriptional repressor promyelocytic leukemia zinc finger protein (PLZF) at the rs12101261 site. Repression by PLZF depended on the rs12101261 disease susceptibility allele and was increased by IFNα. Intrathymic TSHR expression was decreased in individuals homozygous for the rs12101261 disease-associated genotype compared with carriers of the disease-protective allele. Our studies discovered a genetic–epigenetic interaction involving a noncoding SNP in the TSHR gene that regulates thymic TSHR gene expression and facilitates escape of TSHR-reactive T cells from central tolerance, triggering GD.

Graves disease (GD) is an autoimmune condition characterized by infiltration of the thyroid by T and B cells reactive to thyroid antigens, leading to production of thyroid stimulating hormone receptor (TSHR)-stimulating antibodies (TSAbs), which induce biochemical and clinical hyperthyroidism (1). GD arises when a confluence of genetic susceptibility and environmental encounters leads to loss of immune self-tolerance at central and peripheral levels (2). Several GD susceptibility loci have been identified, including HLA-DRβ1-Arg74, CD40, CTLA-4, PTPN22, thyroglobulin (Tg), and the TSHR (2). The TSHR is unique among these susceptibility genes in that it is the target of the autoimmune response in GD. Indeed, TSAbs are the hallmark of the disease and are the direct cause of clinical symptoms (2). Comprehensive sequence analysis of the TSHR gene pinpointed the associated variants to intron 1 (3). However, the causative variant and the mechanisms by which it predisposes to GD are still unknown, and the functionality of the single-nucleotide polymorphisms (SNPs) in intron 1 has not been investigated.

Recently, it was shown that nongenetic triggers of autoimmunity, such as viral infections and cytokines, can promote loss of immune tolerance by modulation of gene expression through epigenetic mechanisms (4). Moreover, epigenetic changes in response to the environment can occur at sites of disease-associated polymorphisms, resulting in profound effects on gene expression through modulation of chromatin accessibility to transcriptional regulators (5). Indeed, we have recently shown that chromatin modifications induced by interferon alpha (IFNα) at a Tg gene variant associated with thyroid autoimmunity resulted in significant changes in Tg expression levels (6).

Because the GD-associated SNPs in the TSHR are restricted to 40 kb of intron 1 (3), we hypothesized that regulatory chromatin modifications induced by cellular triggers (e.g., cytokines) can reveal the causative SNP(s). Here we demonstrate that IFNα, a key cytokine secreted during viral infections previously shown to trigger autoimmunity (7), interacts through chromatin remodeling with an SNP in intron 1 of the TSHR gene to reduce thymic TSHR expression. These results suggest that epigenetic–genetic interactions leading to decreased thymic self-antigen expression may be a general mechanism in autoimmunity.

Results

Effects of IFNα on Histone H3 Methylation and Gene Expression in Thyroid Cells.

To characterize genome-wide changes of histone marks induced by IFNα, we exposed ML-1 human thyroid cells to IFNα and performed chromatin immunoprecipitation (ChIP) sequencing (seq) for histone H3 lysine 4 (K4)-trimethylated (H3K4me3) and K4-monomethylated (H3K4me1). First, we analyzed the distribution of methylation signals relative to five gene annotation sets (SI Appendix, Fig. S1). Treatment with IFNα significantly changed the distribution of H3K4me1 across the annotation sets (P < 0.001) (SI Appendix, Fig. S1). The most significant change was an increase of H3K4me1 signals in intronic and 5′ upstream gene regions in IFNα-treated samples compared with controls (SI Appendix, Results and Fig. S1).

ChIP-seq–generated tags were aligned to human genome reference hg18. H3K4me3 and H3K4me1 peaks with significantly higher read coverage in IFNα-treated compared with untreated cells (P < 1 × 10⁻⁸) were designated “enrichment peaks,” whereas peaks with significantly higher read coverage in untreated cells (P < 1 × 10⁻⁸) were considered “depletion peaks.” For H3K4me3, we identified 22,820 enrichment peaks and 46,350 depletion peaks, and for H3K4me1, 38,699 enrichment peaks and 21,896 depletion peaks (SI Appendix, Table S1). We confirmed the ChIP-seq results for a selection of genes (SI Appendix, Fig. S2 A and B).

Significantly more H3K4me3 enrichment peaks overlapping promoter regions [−1 kb to +1 kb from the transcriptional start site (TSS)] were present in IFNα-treated compared with untreated cells (P < 1 × 10⁻⁸) (Fig. 1A and SI Appendix, Table S1). Comparative heatmaps of H3K4me1 modifications in 5′ regions (−10 kb to +1 kb from the TSS) also showed a higher representation of H3K4me1 enrichment peaks in IFNα-treated cells compared with untreated controls (Fig. 1B and SI Appendix, Table S1).

Fig. 1.

Increased H3K4me3 and H3K4me1 signals in IFNα-stimulated cells. (A) Heatmaps showing the intensity of H3K4me3 signals in regions spanning 1 kb up- and downstream of the TSSs of annotated genes. (B) Heatmaps displaying the intensity of H3K4me1 signals in regions spanning 10 kb up- and 1 kb downstream of the TSSs of known genes. Each row represents a region with statistically significant differences in H3K4me3 and H3K4me1. Black vertical lines represent the TSS. Heatmaps for both enrichment (red) and depletion (blue) are shown in control and IFNα-stimulated cells.

Pathway Analysis.

Gene ontology (GO) analysis showed that the main pathways with clustering of histone hypermethylated genes were pathways of immune and defense responses, innate immune responses, and responses to viral infection (SI Appendix, Table S2).

We next explored the relationship between histone modifications and mRNA expression induced by IFNα. RNA-seq analysis showed 151 up-regulated and 14 down-regulated genes (P ≤ 0.01). Reflecting the stimulation by IFNα, Ingenuity Pathway Analysis (IPA) showed an up-regulation of classical IFNα and immune pathways as the top pathways (SI Appendix, Fig. S3). We also used IPA to identify relationships between genes induced by IFNα. Fig. 2A illustrates known functional relationships among genes that were up-regulated in IFNα-treated cells (Fig. 2A). Many of these genes (e.g., OAS2, MX1, ISG15) are known targets of IFNα. Induction of IFNα signature genes was also associated with IFN regulatory factors and NF-κB. Similarly, GO analysis revealed pathways directly related to innate or adaptive immune responses: (i) response to virus (P < 1 × 10⁻⁸); (ii) immune response (P < 1 × 10⁻⁸); (iii) defense response (P = 5.11 × 10⁻⁷); and (iv) innate immune response (P = 3.78 × 10⁻⁶) (Fig. 2B and SI Appendix, Table S3).

Fig. 2.

IFNα induces up-regulation of gene expression in thyroid cells. (A) IPA network for a subset of genes induced by IFNα. Red, up-regulated genes. For each gene, the log₂ ratio is shown. (B) Top GO pathways of genes induced by IFNα. Gray bars, the number of genes within the GO attribute; black bars, the number of genes with H3K4me3 and H3K4me1 enrichment. BP, biological process.

Nineteen percent of genes up-regulated by IFNα showed increased H3K4me3 in the regions −2.5 kb to +1 kb from the TSS (P < 1 × 10⁻⁸) (SI Appendix, Table S4). Many of these genes were known direct targets of IFNα (SI Appendix, Table S2). Thirty-four percent of the up-regulated genes included in the top GO pathways were associated with either H3K4me3 (−2.5 kb to +1 kb from the TSS) or H3K4me1 (−10 kb to +1 kb from the TSS) (Fig. 2B and SI Appendix, Table S3). These data suggested that IFNα induces gene transcription through induction of epigenetic changes activating markers of open chromatin in cells.

Two Adjacent GD-Associated SNPs Within Intron 1 of the TSHR Display IFNα-Induced Enrichment of H3K4me1.

We aligned 79 SNPs associated with autoimmune thyroid disease (AITD) with identified H3K4me1 peaks. We found two adjacent SNPs in intron 1 of the TSHR gene that overlapped with an H3K4me1 enrichment peak—rs12101261 and rs12101255 (SI Appendix, Table S5). Both rs12101261 (8) and rs12101255 (9) are associated with GD; they are located within 176 bp of each other and are in tight linkage disequilibrium (LD). Confirming the ChIP-seq results, ChIP-quantitative (q)PCR showed that IFNα treatment significantly increased H3K4me1 in the region overlapping rs12101261 and rs12101255 (Fig. 3A). ChIP in primary thyroid cells showed a similar pattern of H3K4me1 enrichment as for ML-1 cells (Fig. 3B). Increased H3K4me1 overlapping the two SNPs in TSHR intron 1 was also found in thyroid tissues removed at surgery (SI Appendix, Fig. S4). As expected, we found a decrease in H3K4me3 in the region spanning rs12101261 and rs12101255 compared with the TSHR promoter (Fig. 3C).

Fig. 3.

H3K4me1 enrichment at rs12101261 and rs12101255. (A) In ML-1 cells, treatment with IFNα-induced H3K4me1 enrichment in the region overlapping the two SNPs (P = 0.04) but not in the TSHR promoter region. Results are presented as means ± SD of five independent experiments. (B) In primary thyroid cells, genomic sequence overlapping the SNPs showed increased H3K4me1 (white bars). IFNα treatment further increased H3K4me1 enrichment compared with the promoter region (black bars). Data is representative of two independent experiments. Error bars reflect triplicate qPCRs. (C) H3K4me3 depletion at rs12101261 and rs12101255. Results are presented as means ± SD of three independent experiments. White bars, control samples treated with H3K4me1 Ab; black bars, samples exposed to IFNα and treated with H3K4me1 Ab; gray bars, samples treated with IgG control Ab.

Identifying the GD Causative SNP in Intron 1 of the TSHR.

Previous studies of the TSHR gene/locus localized the causative variant to a 40-kb region within intron 1 where at least five GD-associated SNPs were identified (8–11) (SI Appendix, Fig. S5A). We computed the LD between these GD-associated SNPs, and found that all of them are in tight LD (D′ = 0.95) (SI Appendix, Fig. S5B), demonstrating that identifying the causative SNP(s) is beyond the resolution of association studies. However, as described above, our epigenomic analysis showing that IFNα causes significant H3K4me1 enrichment only in the region overlapping rs12101255 and rs12101261 strongly suggested that one of them is the causative SNP (SI Appendix, Fig. S5 C and D).

To identify the causative SNP and the mechanism by which it triggers disease, we used the Genomatix MatInspector database (12) to search for predicted transcriptional factor (TF) binding motifs that overlap rs12101255 and rs12101261. Indeed, rs12101261 overlaps a predicted motif for the TF promyelocytic leukemia zinc finger protein (PLZF). A second predicted PLZF motif was mapped 32 bp upstream of rs12101261. No TF motifs were found to overlap rs12101255. Moreover, there was robust expression of PLZF in thyroid and thymus cells (SI Appendix, Fig. S6A).

PLZF Mediates Transcriptional Repression of the TSHR.

To assess the effects of PLZF on TSHR transcription, we used two thyroid cell lines, ML-1 and FTC-133. Both cell lines have the same rs12101261 genotype (CC) and express TSHR mRNA (Fig. 4 A and B) and protein (SI Appendix, Fig. S6B). However, ML-1 cells express PLZF, whereas FTC-133 cells do not (Fig. 4 A and B). After treatment with IFNα, TSHR mRNA expression was down-regulated by 40% in ML-1 cells (P = 0.03) (Fig. 4A) but remained unchanged in FTC-133 cells (Fig. 4B). Down-regulation of TSHR expression by 40% was also found in primary thyroid cells treated with IFNα (SI Appendix, Fig. S7). ChIP analysis revealed an increase of H3K4me1 at the rs12101261 site in ML-1 compared with FTC-133 cells (SI Appendix, Fig. S8). To assess the relationship between PLZF and TSHR expression, we transfected FTC-133 cells with a PLZF-expressing vector and assessed TSHR mRNA levels. Expression of PLZF in FTC-133 cells significantly reduced TSHR levels in a dose-dependent manner (Fig. 4C). Treatment with IFNα augmented this PLZF effect (P = 0.01) (Fig. 4D). PLZF associates with histone deacetylases (HDACs) to mediate gene repression (13). To determine whether HDAC activity was required for repression of TSHR expression, we examined the effect of Trichostatin A (TSA), a specific inhibitor of HDACs, on PLZF activity. TSA significantly reduced the inhibitory effect of PLZF (Fig. 4E).

Fig. 4.

PLZF down-regulates TSHR gene expression. (A and B) TSHR mRNA expression in (A) ML-1 cells and (B) FTC-133 cells treated and untreated with IFNα. IFNα-induced OAS1 mRNA expression was used as control. Results are mean fold change ± SD of seven and four independent experiments for ML-1 and FTC-133 cells, respectively. (C) PLZF overexpression in FTC-133 cells decreases TSHR gene expression. (D) IFNα increased PLZF effects in FTC-133 cells. PLZF and IFNα induction of RSAD2 was used as control. (E) Treatment with TSA restores TSHR expression. Data represent the mean ± SD of four independent experiments. Differences in mRNA levels were compared by the t test for independent samples. *P ≤ 0.05; **P ≤ 0.001; ***P ≤ 0.0001.

Differential Binding of PLZF to the rs12101261 Disease-Associated T Allele vs. the Disease-Protective C Allele.

We tested whether allele variability at rs12101261 generates differences in TSHR promoter activity. The sequence overlapping rs12101261 was placed upstream of the TSHR proximal promoter into a luciferase reporter vector. We then substituted the original disease-protective C allele [Fig. 5A, (b)] with the disease-associated T allele [Fig. 5A, (c)]. Construct (b) containing the disease-protective C allele decreased TSHR promoter activity by 22% (P = 0.04), whereas construct (c) containing the disease-associated T allele reduced the promoter activity by 78% (P = 4 × 10⁻⁶) (Fig. 5A). We next tested the importance of each of the PLZF sites in the context of allele variability at rs12101261. Deletion of the PLZF binding site overlapping rs12101261 decreased promoter activity by 22% (P = 0.004) [Fig. 5A, (d)]. Deletion of the PLZF binding site upstream of rs12101261 when the disease-protective C allele was present had no effect on TSHR promoter activity (P = 0.7) [Fig. 5A, (e)]. The same deletion but in the presence of the disease-associated T allele decreased promoter activity by 22% (P = 0.04) [Fig. 5A, (f)]. These results suggested that both PLZF binding sites contribute to suppression of TSHR promoter activity but that the effect is stronger in the presence of the disease-associated T allele.

Fig. 5.

PLZF regulates TSHR promoter activity. (A) Luciferase reporter analysis of PLZF-dependent TSHR promoter activity. The TSHR proximal promoter (pr.) was cloned into the pGL4.10 luciferase (Luc.) reporter vector to generate construct (a). A sequence harboring rs12101261 containing the C (protective) allele was cloned upstream of the TSHR proximal promoter of construct (a) to generate construct (b). Construct (c) contains the T (susceptible) allele of the rs12101261 SNP. Construct (d) contains an 11-bp deletion of the PLZF binding site overlapping the rs12101261 SNP (PLZF 1). Construct (e) contains the rs12101261 C (protective) allele and an 11-bp deletion of the PLZF binding site upstream of rs12101261 (PLZF 2). Construct (f) contains the rs12101261 T (susceptible) allele and an 11-bp deletion of the PLZF (2) binding site. Each construct was transfected into ML-1 cells and relative luciferase levels were measured. Data represent the mean ± SD of four independent experiments. Statistical differences between different constructs were determined by the t test for independent samples. *P ≤ 0.05; ***P ≤ 0.0001. (B) Knockdown of PLZF by siRNA increases TSHR promoter activity. Constructs (a), (b), and (c) were cotransfected with PLZF siRNA into ML-1 cells and luciferase activities were measured. *P ≤ 0.05; **P ≤ 0.001. (Inset) RT-PCR for PLZF and GAPDH in samples treated with control and PLZF siRNAs. (C) ChIP for PLZF in PBMCs from individuals homozygous for the rs12101261 disease-protective C allele and disease-associated T allele. Data are representative of three independent experiments. (D) PLZF binds strongly to the rs12101261 T allele. HEK-293 cells were cotransfected with constructs (e) and (f) in A and a plasmid expressing PLZF–HaloTag followed by immunoprecipitation using anti-HaloTag Ab. Error bars in C and D reflect triplicate qPCRs.

Next, we cotransfected ML-1 cells with constructs (a), (b), and (c) and with short interfering (si)RNAs targeting PLZF (Fig. 5B, Inset). TSHR promoter activity was unchanged by PLZF siRNAs in the absence of the rs12101261 overlapping sequence [Fig. 5B, (a)]. In contrast, TSHR promoter activity was up-regulated in the presence of either the disease-protective C allele (P = 0.001) [Fig. 5B, (b)] or the disease-associated T allele (P = 0.002) [Fig. 5B, (c)]. These results suggested that PLZF is required for suppression of TSHR promoter activity.

Next, we tested whether allele-specific effects of PLZF were determined by differential binding to the rs12101261 site. We performed ChIP in peripheral blood mononuclear cells (PMBCs) from individuals homozygous for the disease-protective allele (CC) or for the disease-associated allele (TT). Both PLZF and TSHR were expressed in PBMCs (SI Appendix, Fig. S6). We found that PLZF binds 4.6-fold stronger in the presence of the disease-associated genotype than in the presence of the disease-protective genotype at rs12101261 (Fig. 5C). To confirm allele-specific binding of PLZF to the rs12101261 site, we cotransfected HEK-293 cells with a HaloTag vector expressing PLZF and constructs harboring only the PLZF binding site overlapping rs12101261 with either the C allele [Fig. 5A, (e)] or the T allele [Fig. 5A, (f)]. ChIP-qPCR (Fig. 5D) and ChIP-PCR (SI Appendix, Fig. S9) using an anti-HaloTag Ab showed significantly stronger binding of PLZF to the construct containing the disease-associated T allele compared with the disease-protective C allele.

The rs12101261 Disease-Associated T Allele Is Associated with Lower TSHR Expression in Thymus Tissues.

We investigated thymic TSHR mRNA expression in relation to the rs12101261 genotype and PLZF expression levels in thymus tissues from 37 patients who underwent thoracic surgery. The GAPDH-normalized TSHR levels relative to PLZF were compared between individuals carrying different rs12101261 genotypes. Individuals carrying the TT genotype showed 3.4-fold lower median relative expression of TSHR compared with individuals carrying the CC and CT genotypes (P = 0.01) (Fig. 6A). Correlation analysis between TSHR and PLZF levels in individuals carrying the TT, CT, and CC genotypes revealed that only in the group carrying the disease-associated (TT) genotype, PLZF and TSHR expression showed a significant negative linear correlation (Pearson’s r = −0.21), whereas in both the CC and CT groups they showed a positive relationship (Pearson’s r = 0.80 and 0.89, respectively) (Fig. 6B). These results confirmed the significant effect of rs12101261 variation on PLZF-mediated TSHR expression in the thymus.

Fig. 6.

Decreased TSHR mRNA expression levels associated with the rs12101261 T allele in human thymus tissues. (A) Decreased TSHR expression associated with the rs12101261 TT genotype. Horizontal gray bars show the medians of mRNA levels. The median expression levels were 7.09 for the CT + CC genotypes (n = 29) and 2.06 for the TT genotype (n = 8) (*P = 0.01). (B) Correlation between TSHR and PLZF mRNA levels in thymus tissues from individuals who were homozygous TT (n = 9), homozygous CC (n = 11), and heterozygous CT (n = 19) for the rs12101261 SNP.

Discussion

Autoimmune thyroid diseases develop as a result of interacting susceptibility gene variants and environmental triggers. However, it is still unclear how genetic risk variants interact with environmental factors, such as infections, to trigger disease. Recently, it was shown that epigenetic modifications may represent the critical interface for these interactions, and a new model is emerging for the genetic–epigenetic regulation of phenotype. According to this model, certain noncoding SNP alleles facilitate TF binding to regulatory gene regions, thereby leading to histone modifications that up-regulate or down-regulate gene transcription and determine phenotype (14–16). Such genetic–epigenetic regulation of gene expression can increase the risk or even trigger disease in certain individuals.

A major mechanism by which viruses trigger autoimmunity is through local cytokine production, specifically IFNα (17, 18). We hypothesized that IFNα triggers autoimmunity by inducing epigenetic changes at disease susceptibility loci. To test our hypothesis, we analyzed epigenomic and transcriptomic effects of IFNα on human thyroid cells. Most of the genes we found up-regulated by IFNα were immune-response genes, in keeping with our previous study of patients with hepatitis C virus (HCV) infection treated with IFNα that developed thyroiditis (19). Also consistent with these data, we found the same IFNα-induced gene signatures in the thyroids of transgenic mice overexpressing IFNα in the thyroid (17). Together, these data suggest that up-regulation of tissue IFNα (e.g., during viral infections) is associated with unique gene expression patterns that in susceptible individuals can trigger autoimmunity. Indeed, IFNα can trigger several autoimmune diseases, such as systemic lupus erythematosus (20), type 1 diabetes (18), and AITD (21), but its effect is conditioned on genetic susceptibility.

We found that exposure to IFNα induced enrichment of H3K4me1 and H3K4me3 preferentially in noncoding gene regions. Other studies also showed that activation of proinflammatory stimuli was associated with enrichment in H3K4 methylation of inflammatory genes (22, 23). Because the majority of complex disease-associated SNPs were shown to be located in regulatory regions distal to the TSS (24) and these regions are generally marked by H3K4me1, we tested whether the H3K4me1-enriched regions, induced by IFNα, overlapped known AITD susceptibility SNPs. An IFNα-induced H3K4me1 peak overlapped a locus in intron 1 of the TSHR gene that included two adjacent GD-associated SNPs—rs12101255 and rs12101261. Recently, the GD-associated locus was fine-mapped to a 40-kb region in intron 1 of the TSHR overlapping these SNPs (9). SNPs rs12101255 and rs12101261 are only 176 bp apart and in strong LD, making it virtually impossible to differentiate which one is the disease-causing variant. However, by functional/mechanistic studies, we demonstrated that SNP rs12101261 is most likely the causative SNP, because it regulates TSHR gene expression in an allele-dependent manner through the binding of the TF PLZF.

PLZF is a transcriptional repressor that functions through recruiting multiprotein complexes to regulatory gene elements (25) and through chromatin remodeling (13). PLZF can be phosphorylated by IFNα, but this phosphorylation induces PLZF to become a transcriptional activator (26). Besides phosphorylation, IFNα was also shown to increase PLZF expression inside nuclear bodies by facilitating its recruitment by promyelocytic leukemia protein (27). In our study, overexpression of PLZF down-regulated TSHR gene expression, showing that it acts as a transcriptional repressor of TSHR. Supporting this conclusion, we showed that inhibition of HDACs reversed the inhibitory activity of PLZF on the TSHR. Indeed, PLZF’s transcriptional activity requires HDAC-associated enzymatic activity (13, 26).

We identified two adjacent PLZF binding sites in intron 1 of the TSHR that bound PLZF to regulate gene expression in a cooperative manner. It has been shown that simultaneous binding of TFs at two nearby sites is energetically favored, because TFs facilitate each other’s occupancy (28). We also showed that the disease-associated T allele provides a stronger binding site for PLZF at rs12101261 and therefore efficiently represses transcription. Our reporter assays showed a more efficient effect of PLZF when both binding sites were intact and the disease-associated allele was present at rs12101261. These results are consistent with previous data by us and others demonstrating the importance of disease-associated SNPs in noncoding regions in impacting transcriptional mechanisms by interrupting TF binding sites (6, 29) or by disturbing long-range interactions (30).

Our data showed down-regulation of TSHR gene expression by the disease-associated allele of SNP rs12101261 in human thyroid cells. However, the implications to the etiology of GD are most likely due to the effects on TSHR gene expression in the thymus. We found that the disease-predisposing genotype (TT) at the rs12101261 SNP was associated with decreased thymic TSHR mRNA expression, and the decrease in TSHR levels correlated with PLZF levels in the thymus. In agreement with our findings, the susceptible genotype (AA) at rs179247, a TSHR intron 1 SNP in tight LD with rs12101261, was also associated with reduced TSHR thymic expression levels (31). Our data support the concept that tolerance to thyroid autoantigens such as the TSHR is achieved by exposing T lymphocytes to thyroid-specific antigens expressed in the thymus, thereby inducing anergy to these antigens (32). Gradual loss of proper epigenetic interactions due to microenvironmental influences, such as IFNα production during viral infections, would affect the regulation of TSHR gene transcription by genetic variants, resulting in impaired TSHR gene expression. Decreased intrathymic TSHR expression can then facilitate pathogenic T-cell escape from central tolerance and increase the risk of autoimmunity to the TSHR. Indeed, studies have underscored the role of type I interferons in thymic development and T-cell selection (33). Moreover, it was shown that fetal thymocytes are susceptible to viral infections (34), and IFNα treatment in patients with HCV infection is associated with inhibition of intrathymic precursor T-cell proliferation (35). Thus, our data support the concept that genetic–epigenetic regulation of intrathymic self-antigen expression (e.g., TSHR) is key to self-tolerance, and that alterations in these mechanisms can decrease autoantigen expression in the thymus, leading to escape of self-reactive T cells.

In summary, genetic–epigenetic modulation of gene expression has recently emerged as an attractive model explaining how noncoding SNPs may confer risk for complex diseases. According to this model, the key interactions are between transcription factors and allelic variants at noncoding regions that induce histone modifications leading to up- or down-regulation of gene transcription (36). Our data support this model by demonstrating an interaction between the transcription factor PLZF and a susceptibility variant located in intron 1 of the TSHR gene leading to decreased TSHR gene expression. This interaction is augmented in the presence of IFNα, a key cytokine secreted during viral infections. Decreased TSHR gene expression in the thymus can lead to breakdown of tolerance and autoimmunity targeting the TSHR.

Materials and Methods

Anonymous deidentified tissue samples were collected using a protocol approved by the Icahn School of Medicine Institutional Review Board. Written informed consent was obtained for PBMCs collection. Detailed description of tissue processing, cell-culture conditions, SNP genotyping, RNA isolation, qRT-PCR and RT-PCR assays, reporter vector construction and luciferase assays, ChIP, ChIP-seq, and RNA-seq analyses, and siRNA-mediated inhibition of PLZF can be found in SI Appendix, Materials and Methods. Primer sequences for PCR, luciferase constructs, and ChIP experiments are provided in Table S6.