Genetic Risk Factors in Drug‐Induced Liver Injury Due to Isoniazid‐Containing Antituberculosis Drug Regimens

Drug‐induced liver injury (DILI) is a complication of treatment with antituberculosis (TB) drugs, especially in isoniazid (INH)‐containing regimens. To investigate genetic risk factors, we performed a genomewide association study (GWAS) involving anti‐TB DILI cases (55 Indian and 70 European) and controls (1,199 Indian and 10,397 European). Most cases were treated with a standard anti‐TB drug regimen; all received INH. We imputed single nucleotide polymorphism and HLA genotypes and performed trans‐ethnic meta‐analysis on GWAS and candidate gene genotypes. GWAS found one significant association (rs117491755) in Europeans only. For HLA, HLA‐B*52:01 was significant (meta‐analysis odds ratio (OR) 2.67, 95% confidence interval (CI) 1.63–4.37, P = 9.4 × 10−5). For N‐acetyltransferase 2 (NAT2), NAT2*5 frequency was lower in cases (OR 0.69, 95% CI 0.57–0.83, P = 0.01). NAT2*6 and NAT2*7 were more common, with homozygotes for NAT2*6 and/or NAT2*7 enriched among cases (OR 1.89, 95% CI 0.84–4.22, P = 0.004). We conclude HLA genotype makes a small contribution to TB drug‐related DILI and that the NAT2 contribution is complex, but consistent with previous reports when differences in the metabolic effect of NAT2*5 compared with those of NAT2*6 and NAT2*7 are considered.

Up to 20% of patients receiving isoniazid (INH) either as mono or combination therapy for tuberculosis (TB) may develop transient asymptomatic elevation of liver enzymes but this elevation usually resolves without drug discontinuation. 1,2 In a recent large prospective cohort study based in China, 5.4% of patients on anti-TB combination therapy developed drug-induced liver injury (DILI) 3 as defined by the International DILI Expert Working Group. 4 In 16% of these cases, DILI was accompanied by other symptoms of hepatotoxicity, including jaundice, and 5.3% developed acute liver failure. 3 Recurrence of DILI upon retreatment (called positive rechallenge) has been reported to occur in ~ 9-25% with at least one of the anti-TB drugs. 1,2,5,6 The incidence of DILI when INH is combined with rifampicin appears higher than for INH alone, with addition of pyrazinamide increasing the risk further. 7 A number of drug-specific and host-related factors influence the susceptibility of a patient to DILI with anti-TB drugs. 8 Potential mechanisms related to INH DILI have been the most widely investigated. Acetylhydrazine is generally considered to be a key INH metabolite contributing to INH-induced DILI and is produced by N-acetyltransferase 2 (NAT2). It can undergo further metabolism by cytochrome P450 to a toxic metabolite or by NAT2 to the less toxic diacetylhydrazine. It has been suggested that fast acetylators (those with NAT2 activity within the normal range) will form diacetylhydrazine efficiently and therefore levels of both acetylhydrazine and toxic P450 metabolites will be low. 9 It has also been proposed that slow acetylators who lack NAT2 activity may form higher concentrations of the toxic metabolite hydrazine by cleavage of the amide bond on INH to form isonicotinic acid. 10 There are also data to suggest formation of a reactive metabolite directly from INH oxidation, which may contribute to liver toxicity through the formation of protein adducts 11 inducing an inappropriate immune response. 12 Because of the important contribution by acetylation to INH metabolism and the existence of common loss of function polymorphisms in NAT2, the gene encoding the acetylation enzyme, there have been a large number of studies examining these variants as DILI risk factors. At least four different meta-analyses, including large numbers of cases, have concluded that slow acetylators have an increased risk of TB drug DILI with an overall odds ratio (OR) varying from 1.59 to 6.42, although the risk varied depending on the precise genotypic definition of slow acetylation and the population studied. [13][14][15][16] Although many of these studies involved a limited number of patients experiencing only mild liver injury, the NAT2 association has also been observed in a study involving moderate to severe DILI cases only. 17 A clinical trial based in Japan with differential dosing with INH on the basis of NAT2 genotype found a lower incidence of DILI when slow acetylators were given a lower dosing regimen. 18 A genomewide association study (GWAS) involving patients from Thailand recently described a genomewide significant signal for NAT2, suggesting a stronger risk for DILI development in those positive for slow acetylator alleles in line with many of the earlier candidate gene studies. 19 However, two GWAS, which included small numbers of European cases with anti-TB DILI, did not find any genomewide significant association with the NAT2 slow acetylator genotype. 20,21 In addition to NAT2 variants, other candidate genetic risk factors for anti-TB DILI have also been investigated, with reports suggesting that genotypes for genes relevant to anti-TB drug disposition and oxidative stress, such as CYP2E1, 22 SOD2, 23 GST isoforms, 23 carboxylesterase (CES) isoforms, 24 and PXR (NR1I2) 25 may modulate risk. HLA genotype is a strong risk factor for a number of forms of DILI 21 and it has been suggested that HLA class II genotype is relevant to risk of DILI due to anti-TB drugs in an Indian population. 26 However, further studies using either candidate gene or genomewide approaches have failed to confirm a role for HLA genotype in susceptibility to anti-TB drug-related DILI in Europeans. 20,21,27 A recent study of DILI due to anti-TB drugs in an Ethiopian population who were also HIV-positive and receiving anti-HIV treatment found an interesting association with the class I HLA B*57 alleles, 28 but this form of DILI showed distinct phenotypic differences from that normally associated with anti-TB drugs. A non-HLA immunogenetic risk factor for DILI generally in the gene PTPN22 has also been identified recently but its relevance to anti-TB drug-related DILI has not been investigated in detail to date. 29 The aims of the current study were to perform a GWAS together with additional candidate gene studies on a newly recruited group of Indian patients with moderate to severe anti-TB drug DILI and on an enlarged European cohort, which includes some cases studied previously. 17,20,21

Study design
This study combined in a trans-ethnic meta-analysis framework results from GWAS conducted separately in subjects with European and Indian ancestry. The study was conducted according to the Declaration of Helsinki (Hong Kong Amendment) and Good Clinical Practice (European guidelines). All participants provided written informed consent and each study was approved by the appropriate local (Department of Gastroenterology, St. John's Medical College Hospital, Bangalore, India, and Christian Medical College, Vellore, India), national or institutional ethical review boards, as reported previously. 21

European cohort
We analyzed 70 European ancestry DILI cases exposed at least to INH alone or in combination with one or more of rifampicin, pyrazinamide, and ethambutol, collected by the DILIGEN, iDILIC, and DILIN consortia. Of these cases, 43 were treated with INH alone as standard TB prophylaxis and 27 had been treated with any combination of INH and one or more additional anti-TB drugs with 11 ARTICLE of those exposed to all 4 drugs. These cases all form part of a large cohort of DILI cases previously analyzed by GWAS, 29 with some also included in earlier studies, as summarized in Table S1. A subgroup of 12 UK cases (DILIGEN study) had been included in an earlier study involving direct NAT2 genotyping only 17 but subsequently also underwent genomewide genotyping. 20 European ancestry controls (n = 10,397) were used. 29 Clinical characterization of DILI Criteria used for case definition, categorization of DILI pattern as well as grading of severity of DILI were harmonized across all cohorts using previous guidelines. 4 Causality assessment was done by RUCAM score for the DILIGEN and iDILIC cohorts and by both structured expert opinion and RUCAM score for the DILIN cohort, as previously reported. 20,21 DNA preparation and genotyping For the Indian cases and controls, DNA was isolated from whole blood samples using the standard phenol-chloroform method. DNA samples were genotyped in one batch using the Illumina Human Core Exome-24 BeadChip by the Department of Molecular Genetics, Madras Diabetes Research Foundation, India. DNA isolation and genotyping of European cases from the DILIGEN, iDILIC, and DILIN studies was as described previously. 20,21 Population structure and imputation Quality control checks on the initial genotype data were performed as described previously. 29 To assess the extent of population structure of the study cohorts, and derive eigenvectors to account for confounding, we applied principal component analysis on each cohort separately, using the smartPCA program from the EIGENSTRAT package (version 3.0) 30 on the overlapping single nucleotide polymorphisms (SNPs; minor allele frequency > 0.01) across the range of genotyping arrays used for typing cases and controls. We used 1000 Genome Project samples as the reference panel to select cases and controls of white and South Indian ancestries. SNP imputation was performed in batches dividing the samples according to ethnicity and genotyping platforms. For each batch imputation was carried out using Michigan Imputation Server, as described previously. 29

GWAS analysis and meta-analysis
We tested for association of each SNP with DILI, separately in Indian and European GWAS, in a logistic regression framework, under an additive genetic model, with adjustment for the principal components from smartPCA to account for population structure using PLINK version 1.07. 31 No other additional covariates were included in the model because we did not have clinical information for controls. Association summary statistics from the two cohorts were combined using effective sample size weighted Z-score fixed-effects meta-analysis, implemented in METAL. 32 Allelic ORs across the two cohorts were obtained through inverse-variance weighting of effect sizes, with heterogeneity assessed with Cochran's Q statistic, 33 implemented in METAL. We reported only those SNPs that attained, in addition to genomewide significance, nominal evidence of association (P < 0.05) with the same direction of effect on DILI in both GWAS phases (internal validation). Genomewide significance with clinical outcomes was defined using a common threshold of P < 5 × 10 −8 . Because the phenotypes studied are rare, the number of cases analyzed was limited. All detailed analyses and Manhattan plots were produced with R (version 3.0.2, The R Project for Statistical Computing, http://www.r-proje ct.org). Regional plots were drawn by LocusZoom, as described previously. 21

HLA analysis
For each cohort, HLA alleles were inferred using HIBAG 34 using the reference predictor panels specific for the genotyping chip and ancestry provided in the software webpage. To impute Indian samples, we used the provided Asian reference data. 34 In total, we imputed 217 HLA alleles in the overall European cohort and 192 in the Indian cohort. We set the major histocompatibility complex (MHC) regionwide significance P value threshold for the HLA allele association to 2.5 × 10 −4 to correct for multiple testing (Bonferroni correction for 200 predicted HLA alleles). Association tests for each HLA allele and meta-analysis were carried out as reported above. Haplotype analysis was performed by Plink version 1.07, including the most significantly associated HLA alleles.
Candidate gene analysis For analysis of additional candidate genes, we selected four genes previously proposed to have a role in the pathogenesis of INH-related DILI, including CYP2E1, CES2, CES1, and PXR/NR1I2. We extracted all variants belonging to each gene from GnomAD (https://gnomad. broad insti tute.org/). We performed an association analysis by Plink version 1.07. 31 Significance was defined using a Bonferroni threshold of P < 0.003 considering 16 multiple comparisons.

Cases and controls
Clinical characteristics of the 55 DILI cases from India and 70 European cases are summarized in Table 1. The total Indian DILI cohort was enriched in cases with severe liver injury with 22% of all patients progressing to acute liver failure, transplantation, or death. The majority of the European DILI cases met the definition of moderate to severe DILI used in our previous European DILI studies. 21 Nine European patients (13%) suffered liver failure, underwent liver transplant, or died ( Table 1).
As controls for the Indian cases, we used 1,199 ethnically matched individuals (105 patients treated for TB by similar regimens who did not develop DILI and 104 healthy adults recruited for this study, 356 from the 1000 Genome Project phase III dataset, and 634 from BioMe dataset (Figure S1)). As European controls, we used 10,397 ethnically matched individuals ( Figure S2).
A schematic summary of the overall study is provided in Figure 1.

GWAS analysis
The case-control GWAS in the European cohort showed one marker that passed the significance threshold (rs117491755, OR = 4.37, 95% ARTICLE confidence interval (CI) 2.702-7.061, P = 1.8 × 10 −9 ; allele frequency (AF) cases = 0.143; AF controls = 0.037; Figure 2a). The rs117491755 is an intronic SNP in ASTN2. For the Indian cohort, there were no genomewide significant associations between DILI and imputed or genotyped variants and rs117491755 did not pass quality control checks in the imputed dataset ( Figure 2b). Trans-ethnic meta-analysis between the two ethnic groups analyzed 4,900,532 shared markers between the two cohorts. None of the variants in the meta-analysis reached genomewide significance (Figure 2c).
Candidate gene analysis NAT2 genotypes. The frequency of the NAT2 alleles in both cohorts are summarized in Table 3. NAT2 allele frequencies for our cohorts were similar to frequencies reported previously with Indians showing a high frequency of the NAT2*6 allele in the control group compared with Europeans, whereas in the European controls, NAT2*5 was the most common allele. 37 We tested association between each allele and DILI correcting for population stratification ( Table 4). We found that NAT2*5 was the most under-represented NAT2 allele in both European and Indian cases, passing the Bonferroni correction for the number of the predicted NAT2 alleles (P Meta-analysis = 0.01; Table 4). NAT2*6 was significantly enriched in Indian cases (P = 0.01). Severe Indian cases (n = 12) showed a slightly higher frequency for NAT2*6 (AF = 0.54) and lower frequency of NAT2*5 (AF = 0.11) compared with the overall cases, with NAT2*5 statistically significant as a protective factor compared with controls (OR = 0.30, 95% CI 0.08-0.94, P = 0.05). Among Europeans, INH alone cases (n = 43) showed a significantly decreased frequency for NAT2*5 (AF = 0.34, P = 0.04; Table 4). The group of European severe cases (n = 9) was too small to detect significant trends. We then assigned NAT2 genotypes for each individual and classified them in four acetylator status groups. We evaluated if there was an enrichment of any of the groups in our cases compared with controls. We found that ultraslow acetylators (those carrying homozygous or compound heterozygous genotypes for NAT2*6 and NAT2*7) were significantly enriched in cases compared with controls (P Meta-analysis = 0.004; Table 5). This enrichment was not significant in a multivariable regression model for which we combined rapid and intermediate acetylators as the baseline group (P European = 0.08 and P Indian = 0.1). Examining the enrichment of single genotype groups, the only set significantly associated with DILI by meta-analysis was NAT2*6/NAT2*7 (P = 0.002), which was also significant in Europeans alone (P = 0.01; Table S4).
Other candidate gene analysis. We extracted variants located in selected candidate genes relevant to INH metabolism (CYP2E1,

DISCUSSION
Despite a relatively large number of published studies, the genetic basis for susceptibility to DILI due to anti-TB drugs, including INH, remains poorly understood compared with DILI caused by certain other drugs, such as flucloxacillin and amoxicillin-clavulanate. There are a number of reasons for this, including (i) the complexity of the phenotype (both mild and more serious cases of DILI due to anti-TB drugs are common); (ii) the fact that the standard treatment typically involves a combination of four different drugs; and (iii) both TB as a disease and DILI induced by anti-TB drug treatment are more common in developing countries where relevant genetic polymorphisms may show differing frequencies than in Europeans where DILI has been studied more extensively, making worldwide comparisons difficult. Furthermore, in some countries where concomitant infection with TB and HIV is more common, assessment of causality as to whether the DILI is due to the anti-HIV drugs, anti-TB drugs, or both, is often unclear.
Similar to the findings in two earlier GWAS involving Europeans, 20,21 we failed to detect genomewide significance when we undertook a GWAS in the Indian population. This is in contrast to the recent GWAS performed on a Thai population, where genomewide significance was seen for NAT2, 19 but is more consistent with a separate GWAS performed in Ethiopians, which also failed After imputation of additional SNPs, assess association of individual SNPs with DILI by logistic regression, using additive model and adjusting for population structure. Perform separately in both cohorts followed by meta analysis. Impute HLA genotypes from genome-wide genotyping data for cases and controls. 217 alleles were imputed in Europeans and 192 in Indians. Assess association of individual alleles with DILI. Set p for significance at less than 2.5x10 -4 . Perform separately in both cohorts followed by meta analysis.
Extract NAT2 genotypes from imputed GWAS data. Assess association of individual SNPs and acetylator genotypes with DILI in cases and controls. Perform separately in both cohorts followed by meta analysis.
Extract individual SNPs in additional candidate genes from imputed GWAS data. Assess association of SNPs individually with DILI in cases and controls. Set significance of p< 0.003 for 16 multiple comparisons. Perform separately in both cohorts.

ARTICLE
to observe genomewide significant signals, 38 although the Ethiopian patients were also undergoing HIV treatments that can cause DILI. In our enlarged European population, which includes the previously studied cases 20,21 as well as eight new cases, 29 we did see one genomewide significant signal in an intronic SNP in ASTN2. This gene product appears to affect synaptic strength by trafficking and degradation of surface proteins. 39 The relevance to DILI is not immediately clear and this signal was not seen in the Indian population. Without a positive replication, we have to consider this signal as either a false-positive or putative until further patients have been studied.
Although the current study and several previous reports failed to detect strong HLA associations, we found some evidence that a rare HLA-C*12:02 HLA-B*52:01 haplotype, which has been recently reported to be a risk factor for Crohn's disease in Asians, might contribute to risk of anti-TB drug DILI in some individuals. 40 The effect of the haplotype seems to be consistent across both cohorts and the association passes the Bonferroni correction based on the number of imputed alleles. Meta-analysis showed that the effect of the B*52:01 allele alone was more significant than for C*12:02. The number of cases positive for the risk allele was low but importantly both cohorts showed this association and that there is an already reported association for the haplotype with an autoimmune disease. 40 In view of the small number of cases positive for the "at risk" haplotype and the absence of a signal in the INH only cases, it is possible that the HLA signal may reflect DILI induced by one of the other anti-TB drugs, possibly pyrazinamide.
A previous HLA class II typing study in an Indian population reported that absence of HLA-DQA1*01:02 and presence of HLA-DQB1*02:01 were risk factors for DILI due to anti-TB drugs. 26 However, in the current study, the most significant findings for class II were increased frequencies of HLA-DQA1*01:03 and HLA-DQA1*03:01, although we did not observe this in Europeans. We also saw no association with B*57 alleles but, as discussed previously, we consider that this particular association may relate to a combination of anti-HIV and anti-TB drug treatment, 28 which we would not expect to see confirmed in the current study.
The PTPN22 variant rs2476601 has recently been found to be an additional risk factor for some forms of DILI showing HLA associations. 29 We therefore also evaluated the role of this variant in our patients with anti-TB-related DILI, but did not find a significant association. However, the allele frequency of rs2476601 in South Asian populations is consistently lower compared with European populations (1% vs. 10%, as reported in GnomAD) so our ability to detect any association was also limited.
In view of very limited genomewide or near genomewide significant signals in the GWAS, we proceeded with additional candidate gene analysis using the GWAS data, focusing on NAT2, given the extensive literature which has demonstrated that NAT2 genotype and phenotype is a risk factor for INH-induced DILI. Three meta-analyses on NAT2 as a DILI risk factor, which include studies published up to 2017 together with two recent large studies, appear to be the most informative to use for comparison with the current study. [13][14][15]19,41 In general, most previous studies on NAT2 genotype as a risk factor have examined all slow acetylators in comparison with either homozygous wildtypes (who are often now classified as the fast acetylator group without inclusion of heterozygotes) or both homozygous wildtypes and those heterozygous for one variant allele only (the traditional fast acetylator group). In some studies, heterozygotes are designated as intermediate acetylators and were analyzed separately.
There is, however, increasing data available that indicates that NAT2*5, which is common in Europeans, South Asians, and Africans, but not in East Asians, is not a true "slow acetylator" allele with the gene product retaining some enzyme activity, whereas the enzymes encoded by both NAT2*6 and NAT2*7 are associated with no activity. 36,42,43 In line with this, a recent study of South African Zulus examined levels of INH and certain metabolites in relation to NAT2 genotype in patients undergoing treatment with INH and did not find a significant difference in drug and metabolite levels when comparing NAT2*5 homozygotes or heterozygotes with those homozygous for two rapid acetylator alleles. NAT2*6 and NAT2*7 alleles were not detected in this population. 44 One relatively small study of phenotype-genotype relationships in healthy Swedish volunteers used INH for phenotype determination and reported a higher metabolic ratio for NAT2*5 homozygotes compared with NAT2*6 homozygotes. 45 The NAT2 genotype distribution among the DILI cases in the current study indicates a protective effect for the NAT2*5 allele with NAT2*4 "neutral" whereas an increased risk was seen for NAT2*6 with the combined "ultraslow" group also showing a statistically significant increased risk. This is in agreement with a meta-analysis reporting an increased risk for NAT2*6 and NAT2*7. 43 A recent study performed in Singapore also suggests that the risk for INH-related DILI is from NAT2*6 and *7 only. 41 This could also explain the recent genomewide significance reported for NAT2 variants in Thailand 19 because NAT2*5 is rarely seen in this East Asian population. One of the earliest reports on NAT2 genotype as a risk factor for INH-related DILI was performed in Taiwan. 46 This study found a small number of individuals positive for NAT2*5 but hepatotoxicity was seen almost entirely in those carrying at least one NAT2*6 or NAT2*7 allele. The biological   We considered (a) in rapid group *4/*4 genotypes; (b) in intermediate group *4/*5, *4/*6, and *4/*7 genotypes; (c) in slow group *5/*5, *5/*6, and *5/*7 genotypes; (d) in ultra-slow *6/*6, *7/*6, and *7/*7 genotypes.
C, concordant; D, discordant; Freq, allele frequency; HetP, heterogeneity P value; Effect dir, effect direction; OR, odds ratio of a multivariate regression model correcting for population stratification; P, multinomial P value; P m , meta-analysis P value.
basis for this complex association with NAT2 genotype is not completely clear but *6 and *7 carriers may be at increased risk of toxicity due to higher levels of the parent drug undergoing metabolism by alternative routes to toxic intermediates, such as hydrazine or possibly by accumulation of the acetylhydrazine metabolite, which also may be converted to hydrazine. 9 INH-related DILI was reported to be more common among East Asians compared with white Europeans and African Americans in early population studies. 47,48 This could reflect the higher frequency of the NAT2*6 and NAT2*7 alleles in these populations 46 compared with those reported for Europeans, 45 despite the overall average higher acetylation activity seen in East Asians. It also remains possible that the recent GWAS findings reported for a Thai population 19 showing significance for NAT2 are not directly comparable to the current study as the liver enzyme elevation thresholds for participation in that study were considerably lower than in the current study. We also studied four additional genes potentially relevant to INH disposition in detail in both cohorts to see if any evidence for trends toward genomewide significance could be detected. These were chosen on the basis of direct relevance to the INH metabolic pathway 9 and either encode enzymes (CYP2E1, CES1, and CES2) or transcription regulators with a role in regulation of gene expression (PXR/NR1I2, which regulates CES expression). Our findings were entirely negative. We believe this is not too surprising and is generally consistent with reports in the existing literature of no significance or small effects. 24,49 Larger studies might enable the detection of smaller effects than was feasible in the current study.
A limitation of this study and most others on DILI due to anti-TB drugs is that in addition to INH, the other drugs used in treatment, especially pyrazinamide, can also cause DILI. All cases in the Indian cohort were related to combination anti-TB drug therapy, whereas a significant proportion of DILI cases in the European cohort were attributable to INH monotherapy so the two cohorts are not identical in terms of drug treatment. This is an important limitation but the results obtained for the two cohorts, especially for NAT2 genotype and to some extent for HLA genotype, are still comparable. Alternative regimens not involving INH show slightly lower incidence of hepatotoxicity but DILI may still occur 50 and although it may be possible to determine which drug causes DILI in a particular patient, this requires a series of individual drug rechallenges, which is difficult to perform and was not feasible in the current study.
In conclusion, we have obtained some evidence that certain NAT2 alleles increase the risk of DILI in patients receiving INHcontaining anti-TB drug regimens. The two cohorts we studied were large compared with most previous studies with a well-defined phenotype but were still small compared with GWAS on more common diseases where relatively small but significant effects can be detected by use of very large cohorts. There is a need for additional studies involving either larger cases numbers or additional meta analyses to better understand the underlying risk factors for DILI due to INH and other anti-TB drugs.

SUPPORTING INFORMATION
Supplementary information accompanies this paper on the Clinical Pharmacology & Therapeutics website (www.cpt-journal.com).