Genetic determinants of antithyroid drug-induced agranulocytosis by human leukocyte antigen genotyping and genome-wide association study

Abstract

Graves' disease is the leading cause of hyperthyroidism affecting 1.0-1.6% of the population. Antithyroid drugs are the treatment cornerstone, but may cause life-threatening agranulocytosis. Here we conduct a two-stage association study on two separate subject sets (in total 42 agranulocytosis cases and 1,208 Graves' disease controls), using direct human leukocyte antigen genotyping and SNP-based genome-wide association study. We demonstrate HLA-B*38:02 (Armitage trend Pcombined=6.75 × 10(-32)) and HLA-DRB1*08:03 (Pcombined=1.83 × 10(-9)) as independent susceptibility loci. The genome-wide association study identifies the same signals. Estimated odds ratios for these two loci comparing effective allele carriers to non-carriers are 21.48 (95% confidence interval=11.13-41.48) and 6.13 (95% confidence interval=3.28-11.46), respectively. Carrying both HLA-B*38:02 and HLA-DRB1*08:03 increases odds ratio to 48.41 (Pcombined=3.32 × 10(-21), 95% confidence interval=21.66-108.22). Our results could be useful for antithyroid-induced agranulocytosis and potentially for agranulocytosis caused by other chemicals.

Figure 1. Regional plot of association signals and LD patterns at the HLA region.

(Top panel): The x axis shows chromosomal positions. The left y axis shows –log₁₀P values from the GWAS. Colours of the dots indicate the LD relationship between rs17193122 or rs116869525 and their neighbouring SNPs in the 4-Mb region (30–32 Mb for rs17193122 and 32–34 Mb for rs116869525) based on our own data (969 samples). The red dashed line indicates the significance threshold (9.56 × 10⁻⁸). The right y axis shows the recombination rate between SNPs calculated based on Asian-ancestry population data from the 1000 Genomes Project. (Bottom panel): LD map based on r² in the associated regions using genotype results from our own data (969 samples). We construct the plot using the Haploview software version 4.2, and r² (× 100) values are depicted in the diamonds.

Figure 2. Association results from the genome-wide scan.

Manhattan plot shows the genome-wide P values (−log₁₀P value) of the Cochran–Armitage test for trend from 522,980 polymorphic SNPs in 42 TiA cases and 927 GD controls. SNPs on each chromosome were given the same colour. Red line, P=9.56 × 10⁻⁸ (as the genome-wide significance cutoff value of this study).

Figure 3. The ROC curve to discriminate TiA cases (n=42) from GD controls (n=927).

Sensitivity, specificity, positive and negative predictive values of the model were 61.09% (95% CI=45.64–76.42), 92.13% (95% CI=90.46–93.60), 21.67% (95% CI=14.67–30.11) and 98.57% (95% CI=97.68–99.18), respectively.

Figure 4. Specific protein–ligand interactions between thionamide drugs and the HLA proteins identified in this work.

(a) Methimazole-HLA-B*38:02. (b) Propylthiouracil-HLA-B*38:02. (c) Methimazole-HLA-B*38:01. (d) Propylthiouracil-HLA-B*38:01. (e) Methimazole-HLA-DRA/DRB1*08:03. (f) Propylthiouracil-HLA-DRA/DRB1*08:03 complexes. The predicted binding affinities are within the range between −4.48 and −6.36 kcal mol⁻¹.

Figure 5. Surface representation of HLA proteins and their sub-pockets.

(a) HLA-B*38:02. The pocket positions are defined on the basis of the crystal structures of peptide-HLA-B*15:01 complex. The sub-pockets of HLA-B*38:01 are located at similar positions. (b) HLA-DRA/DRB1*08:03. The pocket positions are defined on the basis of the crystal structure of peptide-HLA-DR1 complex.