Genomewide Association Study Confirming the Association of NAT2 with Susceptibility to Antituberculosis Drug-Induced Liver Injury in Thai Patients

Antituberculosis drug-induced liver injury (ATDILI) is a common side effect leading to tuberculosis (TB) treatment disruption. The mechanism of the disease remains poorly understood. We conducted a genomewide association study (GWAS) to investigate all possible genetic factors of ATDILI in Thai patients. This study was carried out in Thai TB patients, including 79 ATDILI cases and 239 tolerant controls from our network hospitals in Thailand.

ABSTRACT

Antituberculosis drug-induced liver injury (ATDILI) is a common side effect leading to tuberculosis (TB) treatment disruption. The mechanism of the disease remains poorly understood. We conducted a genomewide association study (GWAS) to investigate all possible genetic factors of ATDILI in Thai patients. This study was carried out in Thai TB patients, including 79 ATDILI cases and 239 tolerant controls from our network hospitals in Thailand. Nearly 1 million single-nucleotide polymorphisms (SNPs) were genotyped across the whole genome using an Illumina OmniExpress Exome BeadChip array. In the discovery stage, we identified strong association signals on chromosome 8 originating from the N-acetyltransferase (NAT2) region. The A allele of rs1495741, the top SNP in the intergenic region of NAT2 and PSD3 (14 kb from NAT2), was significantly associated with ATDILI (recessive model, odds ratio of 6.01 [95% confidence interval, 3.42 to 10.57]; P = 6.86E−11). This particular SNP was reported as a tag SNP for NAT2 inferred phenotypes. The AA, AG, and GG genotypes represented NAT2 slow acetylators, intermediate acetylators, and fast acetylators, respectively. The tag SNP genotypes demonstrated a concordance rate of 94.98% with NAT2 acetylator phenotypes. This GWAS shows that NAT2 is the most important risk factor for ATDILI in the Thai population.

TEXT

Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis. Worldwide, TB ranks 9th as a cause of death. In 2016, 10.4 million individuals were estimated as developing TB, with 6.3 million new TB cases reported. The worldwide TB incidence and mortality rates are 2% and 3% per year, respectively. The number of HIV-positive TB patients is 476,774. The global treatment success rate is 83%. A total of 129,689 patients were treated for drug-resistant TB in 2015 (1). Thailand, with a population of 69 million, has a high burden of TB cases, with an estimated TB rate of 172 per 100,000 population per year as of 2016. A total of 68,040 new TB cases were reported in Thailand in 2016, and the mortality rate is 13 per 100,000 population (1).

The treatment regimen for TB recommended by the American Thoracic Society, the U.S. Centers for Disease Control and Prevention, and the Infectious Diseases Society of America consists of an intensive 2-month phase of isoniazid (INH), rifampin (RIF), pyrazinamide, and ethambutol and a 4-month continuation phase of INH and RIF (2). However, this regimen frequently causes anti-TB drug-induced liver injury (ATDILI). First-line anti-TB drugs are the most common causes of drug-induced liver injury (DILI). This side effect usually occurs within 2 months of initial treatment. ATDILI leads to early treatment withdrawal, prolonged treatment duration, treatment failure, and emergence of drug resistance (3). According to criteria from an international consensus meeting, ATDILI is defined as an increase in the level of alanine aminotransferase (ALT) or conjugated bilirubin or a combined increase in aspartate aminotransferase (AST), alkaline phosphatase, and total bilirubin levels given that one of these is more than 2 times the upper limit of normal (ULN) (4). The prevalence of liver injury from anti-TB drugs varies from 3% to 20% (5–9). A report from Thailand showed that 6.7% of anti-TB drug users develop ATDILI (10). This report indicated that 64.3% of these ATDILI patients had ALT or AST levels that were over 5 times the ULN without hepatotoxicity symptoms (nausea/vomiting, anorexia, abdominal pain, fatigue, and jaundice), 9.5% of these ATDILI patients had AST or ALT levels that were more than 3 times the ULN to 5 times the ULN with jaundice, and 11.9% of these ATDILI patients had similar levels of AST and ALT without jaundice (10).

In general, the development of DILI is associated with several important risk factors, such as advanced age, gender, malnutrition, complications of disease, alcohol intake, and genetic factors (11). In addition, many candidate gene studies reported associations between ATDILI and N-acetyltransferase 2 (NAT2) (11–13), cytochrome P450 2E1 (CYP2E1) (14), glutathione S-transferase M1 (GSTM1) (15), glutathione S-transferase T1 (GSTT1) (16), superoxide dismutase 1 (SOD1) (17), HLA-DQA1/DQB1 (18), and the solute carrier organic anion transporter family member 1B1 (SLCO1B1) (19).

To date, there has been only one reported genomewide association study (GWAS) of associations with ATDILI, which was conducted in Ethiopians (20). However, the authors of that study did not identify any single-nucleotide polymorphism (SNP) that achieved genomewide significance. The most significant SNP in their study was rs10946737, which, after adjusting for covariates, demonstrated an odds ratio (OR) of 3.4 (95% confidence interval [CI], 2.0 to 5.6; P = 4.1E−06). This SNP was located in the intron of FAM65B (family with sequence similarity 65 member B) on chromosome 6. Four other top SNPs (rs320035, rs393994, rs319952, and rs320003) were located in the intron of AGBL4 (ATP/GTP binding protein-like 4) on chromosome 1. The effects of these two genes on ATDILI are unclear, however, and thus need further investigation (20).

Our previous candidate gene studies showed an association of NAT2 slow-acetylator status with ATDILI (OR = 8.8 [95% CI, 4.0 to 19.31]; P = 1.53E−08) (21). We conducted an extended study of the previous study by Wattanapokayakit et al. using a GWAS approach to identify all possible genetic factors associated with ATDILI in Thai TB patients in order to provide a better understanding of the mechanism of ATDILI.

RESULTS

The 960,919 SNPs identified in 80 ATDILI patients and 242 tolerant controls were genotyped using an Illumina OmniExpress Exome BeadChip array. We utilized principal-component analysis (PCA) to deal with possible population stratification and removed two samples which were separated from the Asian cluster (Fig. 1). After removal, the genomic inflation factor (λ) was 1.00813 (Fig. 2). Subsequently, we applied standard quality control (QC), and 79 ATDILI cases and 239 controls with 545,492 SNPs passed the QC. The demographic data for patients who passed the QC are shown in Table 1 and in Table S1 in the supplemental material. There were no significant differences in age and gender between cases and controls in this data set. All common variants with a minor allele frequency (MAF) of >0.05 were investigated for their associations with ATDILI in three different inheritance models: allelic, dominant, and recessive. The minimal P (P_min) values among these genetic models were plotted as −log₁₀(P_min) against the chromosome position across the genome to construct a Manhattan plot (Fig. 3). A marked signal that achieved genomewide significance was detected on chromosome 8. The regional plot of that particular position showed that all genomewide associated SNPs were located near the NAT2 gene (Fig. 4). We validated all genotyped SNPs with a P value of <1.0E−05 by an invader assay or direct sequencing. Regional imputation using global references from 1000 Genomes Project phase 3 was also carried out to impute SNPs that were not included in the current genotyping platform. The imputation results did not reveal any additional significant signals (Fig. 4). Therefore, the most significant SNP both before and after imputation was rs1495741 (P_min = 6.86E−11; OR_recessive, 6.01 [95% CI, 3.42 to 10.57]) (Table 2). SNPs rs4921914 and rs4921913 were almost in complete linkage disequilibrium (LD) with rs1495741 (r² = 0.99 and 0.98, respectively). A conditional analysis revealed no secondary associated SNP at the same locus. When we extended the threshold P value to <1.0E−05, we identified the suggestive associated SNPs rs2756263 and rs2756262 on chromosome 20 located near PRND (prion-like protein Doppel) (P_min = 6.93E−06; OR_dominant, 4.47 [95% CI, 2.19 to 9.11]) (Table 3). Furthermore, we examined the expression quantitative trait locus (eQTL) of each of these candidate SNPs in the NAT2 locus and found that the T allele of rs4921911 increased the expression of NAT2 in the testis (P = 1.4E−06) but not the liver. Two SNPs located in the 3′ untranslated region of PRND, rs2756262 and rs2756263, were not significant in the eQTL analysis.

FIG 1.

Principal-component analysis (PCA) of principal component 1 (PC1) and PC2 of Thai TB samples. (A) PCA with 7 populations from 1000 Genomes Project phase 3. THA, Thai samples (ATDILI cases and controls); YRI, Africans; KHV, Vietnamese; JPT, Japanese; CHS, Southern Han Chinese; CHB, Han Chinese; CDX, Chinese Dai; CEU, Europeans. (B) PCA with East Asian (EAS) cluster. (C) PCA among Thai cases (THA_CA) and controls (THA_CO). The red block indicates two samples that were removed.

FIG 2.

Quantile-quantile (QQ) plot of expected P values versus observed P values in the chi-square test (λ_GC = 1.00813).

TABLE 1.

Demographic data of patients in this study

Characteristica	Value for group		P value
	ATDILI (79 patients)	Control (239 patients)
No. of male patients/no. of female patients	45/34	150/89	4.32E−01
Median age (yr) ± IQR	52 ± 27	47 ± 24	6.22E−02
Median no. of days to onset of liver injury ± IQR	18 ± 23
Median peak level of liver enzyme ± IQR (IU/liter)
AST	161 ± 233	27 ± 17	<2.20E−16
ALT	128 ± 197	21 ± 22	<2.20E−16
ALP	143 ± 117	70 ± 34	2.00E−15
TBIL	2.2 ± 3.8	0.6 ± 0.4	2.58E−10

^aIQR, interquartile range; AST, aspartate aminotransferase; ALT, alanine aminotransferase; ALP, alkaline phosphatase; TBIL, total bilirubin.

FIG 3.

Manhattan plot of –log₁₀(P) values across the whole genome.

FIG 4.

Regional plots of NAT2 genotyped SNPs (A) and genotyped and imputed SNPs (B). cM/Mb, centimorgan/megabase.

TABLE 2.

SNPs that showed significant associations with ATDILI at genomewide significance levels (P < 5E−08) in the Thai data set^a

Chromosome	SNP	Nearest gene	Risk allele 1/risk allele 2	RAF		Allelic				Dominant (11 + 12 vs 22)				Recessive (11 vs 12 + 22)				Pmin
				Cases (n = 79)	Controls (n = 239)	OR	L95	U95	P value	OR	L95	U95	P value	OR	L95	U95	P value
8	rs1495741	NAT2	A/G	0.84	0.59	3.70	2.32	5.88	2.64E−09	3.50	1.04	11.82	3.17E−02	6.01	3.42	10.57	6.86E−11	6.86E−11
8	rs4921914	NAT2	A/G	0.84	0.59	3.67	2.30	5.83	2.84E−09	3.50	1.04	11.82	3.17E−02	5.89	3.35	10.35	9.37E−11	9.37E−11
8	rs4921913	NAT2	A/G	0.84	0.59	3.50	2.21	5.53	9.03E−09	3.50	1.04	11.82	3.17E−02	5.54	3.17	9.67	4.34E−10	4.34E−10
8	rs10103029	NAT2	A/G	0.84	0.62	3.33	2.09	5.30	5.71E−08	2.09	0.70	6.23	2.52E−01	5.59	3.17	9.86	6.13E−10	6.13E−10
8	rs10088333	NAT2	G/A	0.84	0.62	3.33	2.09	5.30	5.71E−08	2.09	0.70	6.23	2.52E−01	5.59	3.17	9.86	6.13E−10	6.13E−10
8	rs7816847	NAT2	A/C	0.84	0.60	3.48	2.19	5.54	1.36E−08	3.36	0.99	11.37	4.75E−02	5.35	3.05	9.38	1.38E−09	1.38E−09
8	rs12674710	NAT2	A/C	0.84	0.62	3.18	2.01	5.03	1.73E−07	2.09	0.70	6.23	2.52E−01	5.25	2.99	9.20	1.56E−09	1.56E−09
8	rs4646246	NAT2	A/G	0.84	0.62	3.18	2.01	5.03	1.73E−07	3.09	0.91	10.51	7.06E−02	4.75	2.73	8.27	1.04E−08	1.04E−08
8	rs1041983	NAT2	A/G	0.70	0.50	2.33	1.59	3.42	1.33E−05	1.60	0.76	3.35	2.34E−01	4.92	2.83	8.53	1.76E−08	1.76E−08
8	rs4646267	NAT2	A/G	0.84	0.62	3.14	1.99	4.98	1.80E−07	3.24	0.96	10.99	4.78E−02	4.54	2.61	7.91	2.57E−08	2.57E−08

^aThe allelic model was calculated from allele counts of risk alleles compared with nonrisk alleles. The dominant model was calculated from homozygotes and heterozygotes of risk alleles compared with homozygotes of nonrisk alleles. The recessive model was calculated from homozygotes of risk alleles compared with heterozygotes and homozygotes of nonrisk alleles. RAF, risk allele frequency; OR, odds ratio; L95, lower limit of the confidence interval; U95, upper limit of the confidence interval; P_min, minimum P value among allelic, dominant, and recessive models. 1, allele 1 (risk allele); 2, allele 2; 11+12 vs 22, homozygote and heterozygote of allele 1 compared with homozygote of allele 2 (dominant model); 11 vs 12+22, homozygote of allele 1 compared with homozygote and heterozygote of allele 2 (recessive model).

TABLE 3.

SNPs which showed significant associations with ATDILI at the suggestive significance levels (P value of 1E−05) in the Thai data set^a

Chromosome	SNP	Nearest gene	Risk allele 1/risk allele2	RAF		Allelic				Dominant (11 + 12 vs 22)				Recessive (11 vs 12 + 22)				Pmin
				Cases (n = 79)	Controls (n = 239)	OR	L95	U95	P value	OR	L95	U95	P value	OR	L95	U95	P value
8	rs4921911	NAT2	A/G	0.64	0.45	2.17	1.5	3.14	4.96E−05	1.96	0.99	3.86	6.46E−02	4.00	2.27	7.03	2.00E−06	2.00E−06
8	rs10109286	NAT2	A/G	0.79	0.61	2.39	1.56	3.66	3.08E−05	2.29	0.77	6.77	1.81E−01	3.43	2.02	5.82	5.40E−06	5.40E−06
20	rs2756263	PRND	G/A	0.59	0.40	2.23	1.54	3.21	1.99E−05	4.47	2.19	9.11	6.93E−06	2.00	1.12	3.54	2.75E−02	6.93E−06
20	rs2756262	PRND	G/A	0.59	0.40	2.23	1.54	3.21	1.99E−05	4.47	2.19	9.11	6.93E−06	2.00	1.12	3.54	2.75E−02	6.93E−06

^aThe allelic model was calculated from allele counts of risk alleles compared with nonrisk alleles. The dominant model was calculated from homozygotes and heterozygotes of risk alleles compared with homozygotes of nonrisk alleles. The recessive model was calculated from homozygotes of risk alleles compared with heterozygotes and homozygotes of nonrisk alleles. RAF, risk allele frequency; OR, odds ratio; L95, lower limit of the confidence interval; U95, upper limit of the confident interval; P_min, minimum P value among allelic, dominant, and recessive models. 1, allele 1 (risk allele); 2, allele 2; 11+12 vs 22, homozygote and heterozygote of allele 1 compared with homozygote of allele 2 (dominant model); 11 vs 12+22, homozygote of allele 1 compared with homozygote and heterozygote of allele 2 (recessive model).

The SNP haplotypes of rs1041983 (Y94Y), rs1799929 (L161L), rs1799930 (R197Q), and rs1799931 (G286E) were established according to a conventional method to identify NAT2 alleles. The results of individual SNP association analyses are shown in Table S2. The most significant SNP was rs1041983 (P = 1.76E−08; OR, 4.92 [95% CI, 2.83 to 8.53]). After haplotype construction from these four SNPs, the haplotype CCGG (NAT2*4 allele) exhibited the most significant difference between cases and controls (P = 1.73E−08; OR, 0.29 [95% CI, 0.18 to 0.47]), followed by the TCAG haplotype (NAT2*6A) (P = 6.12E−03; OR, 1.70 [95% CI, 1.15 to 2.50]) and the TCGA haplotype (NAT2*7B) (P = 2.38E−02; OR, 1.63 [95% CI, 1.05 to 2.53]), as shown in Table S3. We then reconstructed haplotypes in which the top SNP was incorporated. LD plots between these five SNPs (rs1041983, rs1799929, rs1799930, rs1799931, and rs1495741) in cases and controls are shown in Fig. S1. These five-SNP haplotype associations exhibited similar levels of association with four-SNP haplotypes (Table S4). Additionally, we frequently found the A allele of rs1495741 together with the haplotype representing NAT2*5B, NAT2*6A, and NAT2*7B. The G allele of rs1495741 was often found together with the haplotype of NAT2*4.

A conventional NAT2 inferred-phenotype association analysis was also carried out on this GWAS data set (Tables S5 and S6). The frequencies of fast, intermediate, and slow acetylators were 0.04, 0.28, and 0.68, respectively, in cases and 0.13, 0.59, and 0.28, respectively, in controls. The results demonstrated that NAT2 slow acetylators had the highest risk of developing ATDILI (P = 3.02E−10; OR, 5.55 [95% CI, 3.19 to 9.62]) (Table S6). Interestingly, genotype frequencies of rs1495741 for GG, GA, and AA were 0.04, 0.24, and 0.72, respectively, in cases and 0.12, 0.58, and 0.30, respectively, in controls. The predictive value of the AA genotype of rs1495741 was also determined, with a sensitivity of 72.15% and a specificity of 69.87% for predicting the risk of ATDILI. In comparison, slow-acetylator status as determined by the conventional approach has 68.35% sensitivity and 71.96% specificity for predicting the risk of ATDILI.

DISCUSSION

The present GWAS involved 79 ATDILI cases and 239 controls with 545,492 SNPs. Ours is the first GWAS of ATDILI to confirm the strong association between the NAT2 gene and the risk of ATDILI. However, the associations of suggestive SNPs with ATDILI in a previous Ethiopian GWAS were not replicated in our Thai GWAS (20) (see Table S7 in the supplemental material). This might have happened due to differences in minor allele frequencies of those SNPs in African and in Thai individuals. In our present GWAS, a primary association was found for rs1495741 in the SNP analysis. The A allele of rs1495741 conferred the highest risk of ATDILI. Previous studies reported associations between this SNP and levels of 1-methylurate in the blood (22), risk of bladder cancer (23, 24), and triglyceride levels in the blood (25, 26). The rs1495741 SNP is located on chromosome 8 in the intergenic region, ∼14 kb from the 3′ end of the NAT2 gene. The results of haplotype analyses incorporating this top SNP with conventional NAT2 haplotypes exhibited similar levels of association compared to the conventional NAT2 haplotypes without the top SNP. Additionally, we found that allele A was linked with slow-acetylator haplotypes and that allele G was linked with fast-acetylator haplotypes. Furthermore, the level of association of the top SNP revealed by the GWAS, rs1495741, with the risk of ATDILI in the recessive model and the level of association of the NAT2 slow-acetylator phenotype (conventional approach) with the risk of ATDILI were comparable. Therefore, we hypothesized that the AA, AG, and GG genotypes of rs1495741 correspond to the NAT2 slow-, intermediate-, and fast-acetylator phenotypes, respectively. The concordance rate was 94.98% comparing the genotypes of rs1495741 and NAT2 predicted acetylator phenotypes. This result was consistent with many previous reports (27–32) demonstrating >90% concordance, except for Africans and black Brazilians (Table S8). This tag SNP, rs1495741, was first discovered by Rothman et al. (23), who found associations between NAT2 predicted phenotypes in a GWAS of bladder cancer. Whether this particular SNP, which is located in the intergenic region, affects the activity of the NAT2 enzyme remains unclear. There are no reports describing an association of this SNP with NAT2 expression in any tissue. All studies reported to date attribute this phenomenon to rs1495741 being a tag SNP. Further investigations are thus needed to obtain a better understanding of the effects of this SNP.

NAT2 encodes N-acetyltransferase 2, which is involved in phase II metabolism. NAT2 is highly expressed in the liver and gastrointestinal tract (33), where it acetylates aryl amines and hydrazine compounds using acetyl-CoA as an acetyl donor. NAT2 is involved in the metabolism of several drugs, including amonafide, procainamide, hydralazine, caffeine, clonazepam, isoniazid (INH), metamizole, phenytoin, sulfamethazine, sulfamethoxazole, and sulfasalazine (34). A meta-analysis of several candidate studies demonstrated an association of NAT2 slow acetylators with ATDILI (35, 36). Metabolism of INH is known to generate hydrazine, a toxic metabolite (37). Coadministration of INH with rifampin, a potent CYP3A4 inducer, also increases the production of hydrazine. INH is metabolized via two pathways. In the first pathway, the major fraction is acetylated by N-acetyltransferase to acetylisoniazid, which undergoes further hydrolysis to acetylhydralazine and subsequent oxidation by CYP2E1 to produce hepatotoxic substances. The second pathway involves the direct hydrolysis of INH to hydrazine (38). The mechanism is illustrated in Fig. S2. The hepatotoxicity of hydrazine is the result of interference with mitochondrial function in the liver (39). One hypothesis to explain how decreased NAT2 activity could lead to liver injury holds that reduced NAT2 activity delays the transformation of acetylhydrazine (hepatotoxic) to diacetylhydrazine (nonhepatotoxic), thus contributing to longer exposure to the toxin in the liver (40). Another hypothesis holds that a reduction in the acetylation process leads to metabolism of INH via direct hydrolysis in an alternative pathway that generates more hydrazine (41).

The 72.15% sensitivity for the tag SNP means that of 100 patients with ATDILI, 72 patients might carry the tag SNP. The 69.87% specificity means that of 100 patients without ATDILI, 70 patients will not carry this tag SNP, indicating an activity high enough for clinical practice. Therefore, we propose that this tag SNP can be used to predict the risk of ATDILI in the Thai population. Although both predictive values of NAT2 slow-acetylation status and the tag SNP are comparable, use of the tag SNP would be simpler and more cost-effective than the conventional method.

We also identified a secondary signal showing a suggestive association (P_min = 6.93E−06; OR_dominant, 4.47 [95% CI, 2.19 to 9.11]) in the PRND gene region on chromosome 20. PRND encodes a Doppel protein that is structurally similar to prion proteins, and an association of PRND with Alzheimer’s disease has been reported (42, 43). PRND is primarily expressed in the brain (43); there are no reports of PRND expression in the liver, suggesting that this gene might be a false positive in terms of ATDILI.

Limitations of this study include its small sample size, which negatively affected the statistical power. It is very difficult to identify additional genetic factors with mild or moderate risk. Moreover, this study lacked information regarding causal drugs because all four anti-TB drugs were concurrently administered to patients, and not all cases had rechallenge histories. Although replication samples were not available in our study, NAT2 has been confirmed in many candidate gene studies in several ethnic groups. Larger sample sizes in future studies will be necessary to identify other genetic factors. As this phenotype is a common side effect of anti-TB drugs and commonly results in drug discontinuation on a global basis, an international collaborative GWAS meta-analysis is needed.

Summary.

This GWAS identified NAT2 as the most important risk factor for ATDILI in the Thai population.

MATERIALS AND METHODS

Sample.

In this study, Thai TB patients who received standard anti-TB medication were recruited from ATDILI network hospitals across the country. Approval was obtained from the ethics committees of the Thai Ministry of Public Health and individual collaborating hospitals, the RIKEN Yokohama Institute, and the University of Tokyo. All patients gave informed consent to participate in this study.

In total, 80 ATDILI patients and 242 tolerant controls were included in this GWAS. The inclusion criteria for ATDILI cases were as follows: (i) patients diagnosed with TB who were on the WHO standard anti-TB drug regimen (2HRZE and 4HR [H, isoniazid; R, rifampicin; Z, pyrazinamide; E, ethambutal]) and (ii) patients who developed liver injury after initiation of anti-TB drugs according to international consensus criteria, with an ALT level ≥2 times the ULN or a combined increase in AST or total bilirubin levels plus one symptom of hepatitis (anorexia, fatigue, jaundice, liver enlargement, nausea, vomiting, and dark urine). Inclusion criteria for tolerant controls were as follows: (i) TB patients who completed TB treatment with a cure outcome, (ii) TB patients whose liver function test results were within the normal range at 2-month and 6-month visits, and (iii) TB patients who had good drug compliance according to the directly observed treatment-, short-course-based national TB control program. The exclusion criteria for both cases and controls were (i) TB patients who were HIV or hepatitis A, B, C, or E positive; (ii) patients with severe illnesses for which the survival period was likely <6 months; (iii) patients with a history of habitual alcohol consumption or alcoholic hepatitis; (iv) patients who received concomitant hepatotoxic drugs such as phenytoin, allopurinol, methotrexate, valproate, carbamazepine, or fluconazole; (v) patients with abnormal baseline liver function test results; (vi) patients who had cognitive dysfunction; and (vii) patients who refused to give consent. This study is an extended study of ths one by Wattanapokayakit et al. with additional ATDILI cases and controls and using a genomewide association approach (21).

DNA extraction.

Genomic DNA was extracted from peripheral blood using a QIAamp DNA minikit (Qiagen, Valencia, CA, USA). DNA quality was assessed using a Nanodrop ND-100 instrument.

Genotyping and quality control.

A total of 960,919 SNPs were genotyped among all cases and controls using an Illumina OmniExpress Exome BeadChip, version 1.4 (Illumina, San Diego, CA). We performed standard quality control (QC) to filter out SNPs with a call rate of <99%, those which deviated from Hardy-Weinberg equilibrium (HWE) among controls (P ≤ 1.0E−06), and those with a minor allele frequency (MAF) of <0.05. For sample QC, we corrected for cryptic relatedness by pairwise identity-by-descent estimation (PI_HAT [proportion identical by descent] of >0.25, first- and second-degree relatives) and removed two subjects. Population stratification was evaluated by principal-component analysis (PCA) with populations from the 1000 Genomes Project (East Asians [EAS], Han Chinese from Beijing [CHB], Southern Han Chinese [CHS], Chinese Dai [CDX], Khin from Vietnam [KHV], Japanese from Tokyo [JPT], Europeans and Caucasians from Utah [CEU], and Yoruba from Nigeria [YRI]) using GCTA software (44). Scatter plots were created from eigenvalues of first and second components, and as a result, two outlier samples from the EAS cluster were removed (Fig. 1). Quantile-quantile plots were drawn from expected P values against observed P values (chi-square test) using the qqman package (45). The genomic inflation factor (λ) was calculated to further identify population stratification and eliminate spurious associations. After the QC process, 79 cases and 239 controls with 545,492 SNPs were used for association analyses.

Regional imputation.

We prepared QC data from regions of interest in single chromosomes exhibiting a genotype call rate of >99% and that deviated from HWE (P < 1.0E−06) with a MAF of <0.01. Haplotype estimation (phasing) was performed with SHAPEIT software using 1000 Genomes Project phase 3 reference haplotypes in order to increase the imputation speed (46). Prephased GWAS haplotypes were used to infer missing genotypes (imputation) from 1000 Genomes Project phase 3 references using IMPUTE2 software (47). Allele frequencies were compared between genotype SNPs in the GWAS data set and after imputation. An imputation information score of 0.9 was applied to the QC for imputed SNPs.

Validation study.

All SNPs exhibiting a P value of <1.0E−05 were selected for the validation study. Multiplex PCR followed by an invader assay was utilized (48).

(i) PCR amplification.

The 20-μl PCR mixture contained 2 μl of template DNA (5 ng/μl), 0.4 μl of Tks Gflex DNA polymerase (1.25 U/μl) (TaKaRa Bio, Otsu, Japan), 9.6 μl of 2× Gflex PCR buffer (Mg²⁺, deoxynucleoside triphosphate [dNTP] plus), 0.04 μl of the forward primer (100 pmol/μl), 0.04 μl of the reverse primer (100 pmol/μl), and 7.92 μl of water. PCR was performed using a GeneAmp 9700 PCR system (Applied Biosystems, Foster City, CA, USA). Optimal conditions were an initial denaturation step at 94°C for 2 min, followed by 50 cycles of 98°C for 10 s, 60°C for 15 s, 68°C for 1 min, and a final extension step at 68°C for 5 min.

(ii) Genotyping.

An invader assay was carried out in a 5-μl reaction mixture containing 0.125 μl of invader buffer, 0.125 μl of fluorescence resonance energy transfer (FRET) buffer, 0.125 μl of Cleavase, 0.25 μl of ROX, 0.25 μl of a probe mixture (prepared with 35 μl of allele probe VIC [100 pmol/μl], 35 μl of allele probe 6-carboxyfluorescein [FAM] [100 pmol/μl], 3.5 μl of an invader probe, and 926.5 μl of water), 2.125 μl of water, and 2 μl of the PCR product diluted 1:10. The invader reaction was conducted at 95°C for 5 min, followed by 63°C for 5 min (incubation time increased as needed). The final products were analyzed using a 7900HT-Fast real-time PCR system (Applied Biosystems). SNPs that were not detected in the invader assay were validated using the Sanger sequencing method.

Conventional method for NAT2 genotyping.

Direct sequencing of the NAT2 gene was performed to identify SNPs of interest in NAT2 exon 2. Interesting NAT2 SNPs commonly found in EAS are rs1041983 (282C>T), rs1799929 (481C>T), rs1799930 (590G>A), and rs1799931 (857G>A). PCR was carried out according to a previously reported protocol (27). Haplotypes were reconstructed using PHASE v.2.1 (49, 50) and confirmed again using Haploview 4.2 (51). NAT2 alleles were determined from haplotypes according to the nomenclature system described at http://nat.mbg.duth.gr/Human%20NAT2%20alleles_2013.htm. NAT2*4 and NAT2*13 are fast-acetylator alleles, whereas NAT2*5B, NAT2*6A, and NAT2*7B are slow-acetylator alleles. The homozygotes of NAT2 fast-acetylator alleles are predicted to be NAT2 fast acetylators. In contrast, homozygotes of NAT2 slow-acetylator alleles are predicted to be NAT2 slow acetylators. Combined fast- and slow-acetylator alleles are inferred as intermediate acetylators.

Expression quantitative trait locus analysis.

Correlations with alterations in gene expression of candidate SNPs were assessed in multiple tissues using the GTEx portal database (52).

Statistical analysis.

All association analyses were carried out using Fisher’s exact test in three different genetic models (allelic, dominant, and recessive) using PLINK 1.09 software (53). The lowest P value among those models was utilized. A Manhattan plot was generated using the qqman package (45) in the R environment, version 3.0.3, to visualize overall associations with SNPs. Regional plots of genomewide significant loci were created using Locus zoom (54). The thresholds for genomewide significance and suggestive significance were 5.0E−08 and 1.0E−05, respectively. A conditional analysis was conducted with the primary associated SNP at a given locus. Haplotype analyses were carried out by comparing four-SNP haplotypes in cases and controls using Fisher’s exact test under the additive model.