Redox Imbalance and Oxidative DNA Damage During Isoniazid Treatment of HIV-Associated Tuberculosis: A Clinical and Translational Pharmacokinetic Study

Background The potential for hepatotoxicity during isoniazid-based tuberculosis (TB) treatment presents a major challenge for TB control programs worldwide. We sought to determine whether pharmacokinetic exposures of isoniazid and its metabolites were related to cellular oxidation/reduction status and downstream markers of oxidative DNA damage. Methods We performed intensive pharmacokinetic sampling among isoniazid-treated patients to determine the relative plasma exposures of isoniazid, acetylisoniazid, hydrazine, and acetylhydrazine. Physiologically-based pharmacokinetic modeling was used to estimate liver tissue exposures during a 24-h dosing interval for each compound. We experimentally treated HepG2 cells with isoniazid and metabolites at equimolar concentrations corresponding to these exposures for 7, 14, and 28-day periods, and performed assays related to redox imbalance and oxidative DNA damage at each timepoint. We related a urine marker of oxidative DNA damage to serum isoniazid pharmacokinetic exposures and pharmacogenetics in a clinical study. Results Among isoniazid-treated patients, serum concentrations of hydrazine and isoniazid concentrations were highly correlated. At equimolar concentrations that approximated hepatic tissue exposures during a 24-h dosing interval, hydrazine demonstrated the highest levels of redox imbalance, mitochondrial injury, and oxidative DNA damage over a 28-day treatment period. In a clinical validation study of isoniazid-treated TB patients, peak isoniazid serum concentrations were positively associated with a urine biomarker of oxidative DNA damage. Conclusions Isoniazid and its metabolites share the potential for oxidative cellular damage, with the greatest effects observed for hydrazine. Future studies should investigate the clinical consequences of oxidative stress with regards to clinical episodes of drug induced liver injury during isoniazid treatment.

Background: The potential for hepatotoxicity during isoniazid-based tuberculosis (TB) treatment presents a major challenge for TB control programs worldwide. We sought to determine whether pharmacokinetic exposures of isoniazid and its metabolites were related to cellular oxidation/reduction status and downstream markers of oxidative DNA damage.
Methods: We performed intensive pharmacokinetic sampling among isoniazid-treated patients to determine the relative plasma exposures of isoniazid, acetylisoniazid, hydrazine, and acetylhydrazine. Physiologically-based pharmacokinetic modeling was used to estimate liver tissue exposures during a 24-h dosing interval for each compound. We experimentally treated HepG2 cells with isoniazid and metabolites at equimolar concentrations corresponding to these exposures for 7, 14, and 28-day periods, and performed assays related to redox imbalance and oxidative DNA damage at each timepoint. We related a urine marker of oxidative DNA damage to serum isoniazid pharmacokinetic exposures and pharmacogenetics in a clinical study.
Results: Among isoniazid-treated patients, serum concentrations of hydrazine and isoniazid concentrations were highly correlated. At equimolar concentrations that approximated hepatic tissue exposures during a 24-h dosing interval, hydrazine demonstrated the highest levels of redox imbalance, mitochondrial injury, and oxidative DNA damage over a 28-day treatment period. In a clinical validation study of isoniazidtreated TB patients, peak isoniazid serum concentrations were positively associated with a urine biomarker of oxidative DNA damage.

BACKGROUND
The potential for isoniazid hepatotoxicity during tuberculosis (TB) treatment remains a major challenge for TB control efforts worldwide. In the United States, isoniazid is the second leading cause of drug-induced liver injury, with evidence of underreporting (Hayashi et al., 2015). Aside from the direct effects of patient injury, the potential for hepatotoxicity during isoniazid treatment places additional burdens on TB control programs (Saukkonen et al., 2006), including laboratory monitoring and treatment interruptions, and hampers drug development efforts (Wallis, 2013). The risk of isoniazid-related hepatotoxicity is further increased in the setting of co-infection with human immunodeficiency virus (HIV) (Ungo et al., 1998).
A central challenge to understanding isoniazid hepatotoxicity has been its uncertain relationship with dose size (Hassan et al., 2015). Damage mediated by reactive oxygen species (ROS), as a consequence of imbalance between cellular oxidation and reduction (Sies, 2015), is a key determinant of drug hepatotoxicity (Jones et al., 2010;Roth and Lee, 2017). The imbalance between oxidation and reduction processes overwhelms cellular stores of glutathione, which provides an intracellular sink for ROS directed by the glutathione-s-transferase (GST) family of proteins (Hayes et al., 2005). With its high cellular density of mitochondria, the liver is particularly vulnerable to injury from ROS byproducts, leading to free radical-induced modifications of host proteins, DNA, and lipids (Friedman and Nunnari, 2014).
While there is abundant evidence that isoniazid and its metabolites cause redox imbalance in mammalian cells (Chowdhury et al., 2001;Hassan et al., 2015;Metushi et al., 2016), clinical studies have not yet linked isoniazid parentmetabolite pharmacokinetic exposures with markers of oxidative stress. Hydrazine, an isoniazid metabolite that damages the electron transport chain in mitochondria, has been proposed as a primary driver of isoniazid hepatotoxicity, yet pre-clinical studies have evaluated concentrations higher than those achieved in patients during isoniazid treatment. Establishing the link, if any, between markers of oxidative stress and hepatotoxicity during isoniazid treatment could lead to alternate approaches to monitoring during treatment, personalization of isoniazid dosing, or the development of ameliorating therapeutic strategies for toxicity prevention (Yew et al., 2018). Here, we hypothesized that long-term exposures to isoniazid and its metabolites were related to markers of redox imbalance in hepatocytes, and that hydrazine exposures would drive redox imbalance and oxidative damage in the liver.

METHODS
Clinical Pharmacokinetic Study to Determine Relative Plasma Exposures of Isoniazid, Acetylisoniazid, Hydrazine, and Acetylhydrazine The parent-metabolite clinical pharmacokinetic study was approved by the Institutional Review Board of Rutgers University, Newark (protocol #20170001150). We performed intensive pharmacokinetic sampling for drug and metabolite concentrations in 10 TB patients (4 latent infection patients and 6 active TB patients) treated with isoniazid, in order to identify the relevant metabolite concentrations for subsequent pre-clinical experiments. After an overnight fast, patients were dosed with oral isoniazid (300 mg). A pre-dose blood draw was performed, with additional blood draws at 1, 2, 4, 6, and 8 h after dosing. Following centrifugation, plasma samples were stored at -80C until shipment to Alturas Analytics (Moscow, ID) for analysis. At each timepoint, we measured plasma concentrations of isoniazid, acetylisoniazid, hydrazine, and acetylhydrazine ( Figure 1). Noncompartmental pharmacokinetic analysis was performed to determine area under the concentration-versus-time curve (AUC), extrapolated over a 24-h dosing interval according to the slope defined in the elimination phase.

Physiologically-Based Pharmacokinetic Model of Liver Tissue Exposures of Isoniazid and Its Metabolites
To predict the hepatic tissue exposures of isoniazid, acetylisoniazid, hydrazine, and acetylhydrazine during isoniazid treatment, the whole-body physiologically based pharmacokinetic model with advanced compartmental and transit model in GastroPlus (Ver 9.7) was used (Supplementary Figure 1). All physicochemical properties for those substances were predicted based on chemical structure from a built-in predictor in GastroPlus. Populationdependent physiological parameters, including tissue volumes and perfusion rate in humans, were obtained using the Population Estimates for Age-Related Physiology module in GastroPlus (Supplementary Table 1). After simulating time-versusconcentration profiles of substances in liver and plasma, noncompartmental pharmacokinetic analysis was performed to calculate the AUC 0-24 ratio between liver and plasma. This ratio was used to predict liver tissue exposures to isoniazid and metabolites, during a 24-h dosing interval, based on the observed plasma exposures in the clinical study.

In Vitro Pharmacologic Exposures of Hepatocytes to Isoniazid and Its Metabolites
The human liver carcinoma cell line, HepG2, is highly differentiated and is often used to screen the cytotoxicity potential of new chemical entities early in drug development (Gerets et al., 2012). HepG2 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin, 100 mg/ml streptomycin, and 2mM L-glutamine, unless otherwise stated. Cells were continuously incubated at 37°C in a humidified 5% CO2/95% air environment and plated in a 12-well plate at 4x10 5 cells/well, over a duration of 24 h prior to compound exposure. Compound stocks dissolved in water were freshly made each exposure day. Cellular supernatant was replaced with 2 ml of culture media containing 10 µM of compound and left to incubate until 90% confluency was reached. Cell passages were conducted every 3 days under continual exposure to the designated compound for the indicated time period. Cellular supernatant was collected at each passage, clarified, aliquoted, and stored at -80°C for future analysis. Treated cells and supernatants were then examined for markers of cytotoxicity, mitochondrial toxicity, and oxidative DNA damage as described below.

Free Thiol Availability Assay
By serving as targets of ROS, thiols provide the cellular defense system against oxidative stress. Glutathione is the principal nonprotein thiol in most cells, and high intracellular glutathione concentrations are required to maintain protein thiols in reduced states, which is essential for cellular functions in metabolism, signaling, and detoxification (Tomanek, 2015). Cellular glutathione depletion was quantified in a free thiol availability assay (Mandavilli and Janes, 2010). Black 96-well polystyrene clear bottom plates were coated with 50 µg/ml of poly-d-lysine for 1 h at 37°C. Following multiple washes with PBS, HepG2 cells from the long-term low-exposure culture growth, at the indicated times, were plated at 3.3x10 5 cells per well and left to incubated at 37°C for 12 h maintain compound exposure. ThiolTracker ™ Violet dye (ThermoFisher) was used to assess free thiol availability following the manufacturer's instructions. Briefly, cells were washed two times with D-PBS +C/M prior to the addition of 20 µM of ThiolTracker. Following a 30 min incubation at 37°C, cells were washed three times with D-PBS +C/M. Fluorescent intensity at 404nm Ex /526nm Em was measured on a Synergy H1 Hybrid Multi-Mode Reader (BioTek).

Mitochondrial Membrane Integrity
To assess mitochondrial membrane integrity, black 96-well polystyrene clear bottom plates were coated with 50 µg/ml of poly-d-lysine for 1 h at 37°C. Following multiple washes with PBS, HepG2 cells from the long-term low-exposure culture growth, at the indicated times, were plated at 3.3x10 5 cells per well and left to incubated at 37°C for 12 h maintaining compound exposure. Tetramethylrhodamine, methyl ester (ThermoFisher, Waltham, Massachusetts) was added at 100 nM at 37°C for 30 min. Following three washes with PBS, fluorescent intensity at 488nm Ex /570nm Em was measured on a Synergy H1 Hybrid Multi-Mode Reader (BioTek).

Cellular and Mitochondrial Toxicity
Mitochondrial toxicity analysis was assayed using the Mitochondrial ToxGlo Assay (Promega, Madison, WI) following the manufacturer's instructions. Briefly, cells obtained from the long-term low-exposure culture growth, at the indicated times, were grown in a 96-well flat clear bottom black polystyrene plate at 3.3x10 5 cells per well in glucose-free (galactose supplemented) culture media. Compound concentrations (10 µM) were consistently maintained throughout the experiment. Following a 24-h incubation with compound, cytotoxicity was evaluated by the addition of a fluorogenic peptide substrate, bis-AAF-R110, to the cell supernatant for 30 min at 37°C. The bis-AAF-R110 substrate penetrates compromised plasma membranes and is activated via interactions with necrosis-associated proteases. After the cells were allowed to incubate for 30 min at 37°C, fluorescence intensity was measured at 485nm Ex /525nm Em . To evaluate mitochondrial toxicity, the plate was acclimated to room temperature for 5 min, at which point the ATP detection reagent was added. Cell lysis mixture was transferred to a 96well flat solid white bottom plate and luminescence intensity was measured. Assays of fluorescence and luminescence activity were performed with a Synergy H1 Hybrid Multi-Mode Reader (BioTek, Winooski, VT).

Comet Chip Imaging of Cellular DNA Damage
Cells treated for 1 week with 10 µM of compound were plated onto a CometChip (Trevigen) at 2x10 5 cells/ml in a single cell suspension (Ge et al., 2014). Cells were incubated on the chip for 60 min at 37°C. The plate was then gently washed with PBS and coated with 1% low melting agarose. After the agarose was solidified by cooling at 4°C, the plate was washed with prechilled non-activated alkaline lysis buffer (2.5 M NaCl, 100nM Na 2 EDTA, 10mM Tris, pH 10). The chip was lysed overnight at 4°C by submerging the chip in the alkaline lysis buffer containing 1% Triton-X. The plate was then washed with PBS and placed in an electrophoresis chamber filled with pre-chilled alkaline electrophoresis buffer (2mM Na 2 EDTA, 300 mM NaOH). Following a 40 min incubation at 4°C to allow for alkaline unwinding, electrophoresis was carried out at 80 volts for 30 min at 4°C. The chip was then neutralized with two incubations at 4°C in 400 mM Tris, pH 7.5 buffer, and equilibrated in 20 mM Tris, pH 7.4 at 4°C for 20 min. DNA staining was achieved by incubating the chip in 100 ml of 0.2X SYBR Gold overnight at 4°C. Fluorescent images were acquired on an Evos FL Imaging System (ThermoFisher).

Immunofluorescence for g-H2Ax
H2Ax is a core histone protein found in the nucleosome, and the phosphorylation of H2Ax is a marker of DNA damage caused by ROS. Variant histone H2AX (g-H2AX) is a marker of DNA doublestrand breaks (Rothkamm et al., 2015). Exposed cells (1 week with 10 µM of compound) were plated in a 12-well cell culture treated plate at 1x10 5 cells/well. Following 24 h incubation at 37°C in media containing 10 µM of compound, cells were fixed with a cold 1:1 solution of methanol and acetone for 20 min at -20°C. The cells were subsequently washed with PBS and blocked for 2 h at room temperature with 1 ml of BlockAid ™ Blocking Solution (Thermo Scientific). Each well was stained with a 1:1,000 dilution of rabbit anti-phospho-H2AFX (Millipore Sigma, St. Louis, MO) in PBS overnight at 4°C with gently rocking. Following multiple washing steps with PBS, cells were incubated with 5 µg/ml of anti-rabbit Alexa Fluor 647-conjugated antibody (ThermoFisher) for 1 h at room temperature. Cells were subsequently washed with PBS and stained with 1 µg/ml of Hoechst 33342 (ThermoFisher) for 5 min at room temperature. Fluorescent images were acquired on an Evos FL Imaging System (ThermoFisher).

8-Hydroxy-2'-Deoxyguanosine (8-OHdG) in Cellular Supernatant
Determination of 8-OHdG content in cellular supernatant was assed using an in-house competition ELISA. Clear 96-well plates were coated with 8-Hydroxy-2'-deoxyguanosine (BioVision, Milpitas, CA) overnight at 4°C. Plates were then washed three times with PBS. Cell supernatant in PBS or standard 8-OHdG standard curve dilutions were added to the appropriate wells. Rabbit anti-8-OHdG (Abcam, Cambridge, MA) was subsequently added at a 1:2,500 final dilution in PBS and the plate was incubated at room temperature with shaking for 2 h. Following three washes with PBS, anti-rabbit HRP antibody was added at a 1:5,000 dilution and the plate was incubated at room temperature with shaking for 1 h. Immediately following five rapid PBS washes, 3.7mM o-phenylenediamine in 0.05 M phosphate-citrate buffer containing 0.03% sodium perborate was added and the plate was incubated at room temperature protected from light for 30 min. Absorbance at 405 nm was then measured on a Synergy H1 Hybrid Multi-Mode Reader (BioTek).

Study Setting and Subjects
We conducted a prospective cohort study of isoniazid pharmacokinetics among HIV/TB patients at 22 public clinics and Princess Marina Hospital in Gaborone, Botswana. The study population has been previously described (Vinnard et al., 2017). Eligibility criteria included citizens of Botswana, HIV infected adults (21 years of age or older) naïve to ART therapy, and newly diagnosed with pulmonary TB initiated on standard first-line TB regimens at weight-based dosing bands recommended in accordance with WHO guidelines. The study visit occurred between 5 and 28 days after initiation of first-line TB therapy that included isoniazid, prior to initiation of ART therapy.

Data Collection
After an overnight fast, the anti-TB drugs were directly administered. Blood samples (10 ml) were drawn at 0, 0.3, 0.9, 2.2, 4.5, and 8 h postdosing, based on optimal sampling theory to estimate isoniazid pharmacokinetic parameters. Plasma isoniazid concentrations were measured with liquid chromatography-tandem mass spectrometry (LC-MS/MS) methods as previously described. The peak plasma concentration (C max ) was obtained directly from the concentrationversus-time data), model-predicted isoniazid concentrations were used to estimate the area under the concentration-versus-time curve during a 24-h interval (AUC 0-24 ). A spot urine sample (approximately 50 ml) was collected 4 h following anti-TB drug dosing and frozen at -80C until analysis. We measured urinary concentrations of 8-OHdG, a marker of oxidative DNA damage that is stable in cryopreserved urine samples, normalized to urine creatinine. The DNA Damage Competitive ELISA kit (ThermoFisher Scientific) was performed using a 1:4 dilution of urine-to-assay buffer. The manufacturer's instructions were followed with one modification to the protocol. Antigen binding was performed overnight at 4°C with shaking in lieu of a 2-h incubation at room temperature. Each assay was performed in triplicate. Optical absorption was assessed using a Synergy H1 Hybrid Multimode Reader (BioTek). To obtain creatinine concentrations for all urine, the Creatinine Urinary Detection Kit (ThermoFisher Scientific) was performed according to the manufacturer's instructions using a 1:20 dilution of urine-todeionized water.
Pharmacogenetic Genotyping for NAT2 and GST Family Genes DNA from whole blood samples was used to perform whole exome sequencing as previously described (Vinnard et al., 2017). Individuals with any combinations of the following NAT-2 alleles: 2*4, 2*11, 2*12, and 2*13, were classified rapid acetylators, while those with both combination of these alleles: 2*5, 2*6, 2*7, and 2*14, were considered slow acetylators, in accordance with existing literature (Doll and Hein, 2001). Patients who possessed one allele from the former group and another from the later were classified as intermediate acetylators. Patients whose single-nucleotide polymorphism (SNP) calls were not available in public databases (http://nat.mbg.duth.gr/ and http://NAT-2pred.rit.albany.edu/) or published literature were designated as ambiguous. We evaluated allele frequencies in SNPs corresponding to GST family genes based on previously identified clinical associations with oxidation/reduction status or isoniazid hepatotoxicity, including GSTA2, GSTP1, and GSTM1 (Ning et al., 2004;Wu et al., 2016). Based on a previously published classification scheme (Tetlow et al., 2001), we categorized GSTA2 haplotypes according to four SNPs (rs2180314, rs6577, rs2234951, rs1803682), as shown in Supplementary Table 3.

Pharmacokinetic Exposures to Isoniazid and Its Metabolites in TB Patients
We obtained parent-metabolite pharmacokinetic data from 10 isoniazid-treated patients, including 4 patients with latent TB infection and 6 patients with active TB disease. The properties of the HPLC-MS/MS plasma assays for isoniazid and metabolites are included in Supplementary Table 2. The spaghetti plots of pharmacokinetic profiles are shown in Figure 2 for isoniazid ( Figure 2A), acetylisoniazid ( Figure 2B), hydrazine ( Figure 2C), and acetylhydrazine ( Figure 2D). For isoniazid, acetylisoniazid, and acetylhydrazine, all concentrations were greater than the lower limit of quantification, corresponding to 1 ng/ml, 10 ng/ml, and 10 ng/ml for isoniazid, acetylisoniazid, and acetylhydrazine, respectively. For hydrazine, 6 of 60 concentrations (10%) were below the lower limit of quantification, corresponding to 2 ng/ml. As expected, plasma AUC 0-24 exposures for isoniazid and hydrazine were highly correlated (R 2 = 0.78), consistent with the role of NAT2 enzyme in the metabolism of each compound (Figure 1).
The physiologically based pharmacokinetic model of isoniazid and metabolite distribution into tissue is included in Supplementary Figure 1. Based on this model, we predicted ratios of liver tissue:plasma exposures for isoniazid, acetylisoniazid, hydrazine, and acetylhydrazine as shown in Table 1. The model predicted an 18-fold higher liver tissue exposure for hydrazine compared to plasma, but lower liver tissue exposures for parent compound and the other metabolites compared to plasma. Based on these model-predicted liver tissue exposures, we selected the concentration of 10 uM to study in preclinical toxicodynamic models for all compounds, corresponding to an AUC 0-24 of 240 uM*hr. This molar concentration fell within 1 log-10 for the median exposure to each compound.

Biochemical Assays of Redox Imbalance and Mitochondrial Injury
There was no appreciable cytotoxicity and of isoniazid or its metabolites ( Figure 3A). However, we observed thiol depletion in treated cells over the 28-day exposure period, demonstrating impaired cellular capacity to maintain the oxidation-reduction equilibrium, greatest among hydrazine-treated cells ( Figure 3B). TMRM is a cell-permeant dye that accumulates in active mitochondria with intact membrane potential and is depleted with free radical injury to mitochondrial membrane potential. Figure 3C shows that there was TMRM depletion in all treated cells across 28 days of exposure, with the greatest depletion observed for hydrazine. Furthermore, Figure 3D shows that cellular ATP production was impaired at 28 days of treatment across all experiments, again with the greatest impairment observed for hydrazine as compared with the other compounds.

Microscopic and Biochemical Assays of Oxidative DNA Damage
We examined levels of oxidative DNA damage in treated HepG2 cells in several ways. First, CometChip microscopy demonstrated greater levels of DNA damage in cells treated for 1 week with hydrazine, acetylhydrazine, and acetylisoniazid, as compared to isoniazid ( Figures 4A, B). Next, we measured g-H2AX in treated cells, a variant histone which corresponds to the presence of double-stranded DNA breaks. Following a 1-week exposure period, g-H2AX levels were greatest during treatment with hydrazine, as compared to isoniazid, acetylisoniazid, and acetylhydrazine ( Figure 4C). Finally, we examined cellular supernatant levels of 8-OhDG, a measure of oxidative DNA damage that is stable in cryopreserved clinical samples (Frijhoff et al., 2015). Figure 4D shows that the highest increase in 8-OhDG levels in cell supernatant was with hydrazine treatment. Given that the 10 µM concentration used in these experiments was less than predicted in liver tissue for hydrazine, but greater than predicted for isoniazid and non-hydrazine metabolites, these findings collectively supported our hypothesis that hydrazine is the primary mediator of redox imbalance in hepatocytes.

Oxidative DNA Damage Among Isoniazid-Treated TB Patients
To further test the clinical relevance of these in vitro findings, we collected urine samples obtained in a previously conducted clinical pharmacokinetic study of HIV/TB patients (Vinnard et al., 2017), to test for 8-OhDG concentrations, a stable marker of oxidative DNA damage, which has been linked to exposures to toxins such as cadmium (Pizzino et al., 2014), chromium, and lead (Pawlas et al., 2017) in other patient cohorts. Demographic, clinical, and immunologic characteristics of study participants are shown in Table 2, which shows that none of the characteristics differed NAT2 genotype. However, NAT2 genotype was associated with urine 8-OhDG concentrations, as shown in Figure 5A. In unadjusted analysis, we observed a significant increase in urine oxidative DNA damage among patients with slow NAT2 genotypes, compared to rapid acetylator genotypes (p=0.04). Next, we examined GSTA2, which encodes the dominant GST protein in human livers, and demonstrates population variability that corresponds to varying levels of hepatic GST expression (Ning et al., 2004). The rs2234951 and rs1803682 variants were not detected in the study cohort, while the SNPs rs6577 and rs2180314 were in complete linkage disequilibrium. Accordingly, all haplotypes were either GSTA2*B or *C under the naming schema proposed by Tetlow et al (Tetlow et al., 2001) Table 3). We identified 25 patients with the homozygous  GSTA2*B/*B haplotype (rs6577/rs6577), 12 patients with the GSTA2*C/*B haplotype (rs6577/rs2180314), and 2 patients with the homozygous GSTA2*C/*C haplotype (rs2180314/rs2180134). In unadjusted analysis, GSTA2 genotype (GSTA2*C/*B or GSTA2*C/*C versus GSTA2*B/*B) was associated with urine oxidative DNA damage, as shown in Figure 5B (p=0.03). In contrast, neither the GSTP1 rs1695 variant, previously identified as a predictor of isoniazid toxicity in a Chinese cohort (Wu et al., 2016), nor the GSTM1 null variant were associated with urine oxidative DNA damage; however the number of patients with the GSTM1 null variant was small (n=3).

(Supplementary
Next, we examined the relationship between isoniazid pharmacokinetic exposures and urine oxidative DNA damage in a multivariate linear regression model. Based on our a priori criteria for model inclusion (greater than 20% change in the regression coefficient when the confounder was included in the model), the final multiple linear regression model included the confounding effects of body weight and renal function on the association between isoniazid C max and urine oxidative DNA damage (Table 3).
Although not a significant confounder of this relationship, we included the effects of GSTA2 genotype (*B/*B vs *C/*B or *C/ *C) based on its independent association with urine levels of 2.5x10 4 (1.6x10 3 -7.0x10 6 ) 1.1x10 5 (86-5.1x10 6 ) 1.3x10 5 (4.3x10 3 -1. oxidative DNA damage in unadjusted analysis. The corresponding increase in adjusted R-squared with sequential variable inclusion, along with the decline in Akaike information criteria (AIC) score, provided statistical support for choosing model as variables were added. Variance inflation factors for the variables in the final model were 1.31, 2.38, 1.78, and 1.39 for isoniazid C max , body weight, creatinine clearance, and GSTA2 genotype, respectively, supporting the absence of significant co-linearity among model predictors. Figure 5C is a marginal regression plot of isoniazid C max and urine oxidative DNA damage, with effect modification by GSTA2 genotype. Although the graphical relationship suggests that GSTA2 genotype acts both as an independent predictor as well as an effect modifier of the relationship between isoniazid C max and urine oxidative DNA damage, the difference in slopes did not reach the a priori threshold for statistical significance (p=0.12 for interaction term).

DISCUSSION
Despite being a cornerstone of TB therapy for 60 years, there is considerable uncertainty regarding the mechanism for isoniazid hepatotoxicity, and the host factors that place TB patients at greatest risk. We first characterized plasma exposures to isoniazid and its metabolites (acetylisoniazid, hydrazine, and acetylhydrazine) in a clinical pharmacokinetic study, demonstrating that isoniazid and hydrazine plasma exposures are highly correlated. Based on these clinical data, we predicted liver tissue exposures using physiologically based pharmacokinetic modeling, observing that hydrazine concentrates in liver relative to plasma. We then developed a long-term, low-exposure experimental set-up in HepG2 cells in order to replicate parent-metabolite pharmacologic exposures in the liver during the course of isoniazid treatment. We found that the greatest levels of mitochondrial injury and oxidative DNA damage occurred with hydrazine treatment at equimolar concentrations. Finally, in a clinical translation of this observation, we observed that peak isoniazid serum concentrations and GSTA2 genotype were independently associated with increasing urine levels of oxidative DNA damage.
These findings link pharmacokinetic exposures to hydrazine, an isoniazid metabolite, with downstream effects on redox imbalance, adding to prior clinical studies of predictors of isoniazid hepatotoxicity. Slow acetylators of isoniazid, as defined by NAT2 genotype, have an increased risk of hepatotoxicity compared to intermediate or rapid acetylators, possessing one or two rapid NAT2 alleles, respectively (Wang et al., 2012). In a prior clinical study, peak serum isoniazid concentrations were directly correlated with elevations in liver transaminases, a marker of hepatic inflammation (Cojutti et al., 2016). Future work should identify whether early markers of oxidative stress among isoniazid-treated patients, such as the urine measure of DNA damage identified in this clinical study, are predictive of subsequent hepatotoxicity events. For example, a cross-sectional study among isoniazid-treated TB patients in India demonstrated an association between hepatotoxicity events and co-incident measurements of plasma oxidative stress markers (Chowdhury et al., 2001). The GSTA2 gene encodes a protein sub-unit of a-GST, the predominant GST enzyme expressed in human liver tissue (Tetlow et al., 2001). Previously published data demonstrated decreased a-GST expression in liver tissue among individuals either heterozygous or homozygous for the GSTA2*C allele, suggesting greater vulnerability to oxidative hepatic injury (Ning et al., 2004). Consistent with this observation, we found that the GSTA2*C haplotype, either heterozygous or homozygous, was an independent predictor of urine oxidative DNA damage among HIV/TB patients. In contrast, we observed no relationship between urine oxidative DNA damage and the rs1695 allele in GSTP1, previously linked to anti-TB drug toxicity among Chinese patients (Wu et al., 2016), or the null mutant GSTM1.
The involvement of glutathione pathways in the development of isoniazid hepatotoxicity could identify both preventive and therapeutic strategies. Hydrazine cellular effects include inhibition of complex II of the electron transport chain, leading to accumulation of ROS that overwhelms cellular glutathione stores, and directly damaging to DNA through free radical injury (Runge-Morris et al., 1994). N-acetylcysteine, clinically used to counter acetaminophen toxicity, has also been used for glutathione deficiency in a wide range of conditions (Atkuri et al., 2007). Identification of intermediary signals on the pathway towards clinical hepatotoxicity events, such as oxidative DNA damage, would also improve efficiency of future clinical trial design of novel TB drug therapies, as isoniazid may be included in these investigational drug combinations (Wallis, 2013).
Aside from hepatotoxicity, oxidative DNA damage during isoniazid treatment may have additional downstream consequences. DNA damage signals induce the nuclear activity of poly(ADP-ribose) polymerase-1 (PARP-1), which is involved in DNA repair activities through multiple pathways (Krishnakumar and Kraus, 2010). Chronic PARP-1 activation competes for intracellular NAD+, resulting in impaired mitochondrial ATP production and cellular apoptosis (Fang et al., 2014). Thus, when considering the clinical implications of increased markers of oxidative DNA damage, additional effects aside from hepatotoxicity events should be considered. For example, isoniazid treatment has been linked to increased markers of cellular apoptosis in immune cells, and direct effects of isoniazid on the patient's immune-metabolism has been suggested. Intriguingly, isoniazid treatment in mice induced apoptosis markers in CD4+ T cells (Tousif et al., 2014). Future clinical studies should directly compare isoniazid and metabolite pharmacokinetic exposures with toxicodynamic markers of redox imbalance, including urine markers of oxidative DNA damage.
This clinical and translational study of redox imbalance related to isoniazid and its metabolites had several important limitations. Multiple overlapping mechanisms may explain the observed relationship between isoniazid metabolites and oxidative stress markers; for example, one possibility is direct inhibition of the thioredoxin system of mitochondria, which is key contributor to oxidation/reduction balance (Stanley et al., 2011). Rather than directly measure concentrations in liver tissue, we used physiologically based pharmacokinetic models to estimate liver tissue exposures for each compound. In the clinical validation study, the metabolite concentrations were not measured, and thus we cannot directly evaluate the relationship between metabolite exposures (such as hydrazine) and oxidative stress in this clinical cohort. Based on our pre-clinical observations, we hypothesize that hydrazine exposures, rather than isoniazid exposures, would demonstrate the greatest association with urine oxidative DNA damage among isoniazid-treated TB patients. Furthermore, we did not perform prospective serial monitoring of hepatic transaminases in the clinical cohort, which was limited to 39 TB patients coinfected with HIV, and the significance of increased urine oxidative stress markers during isoniazid treatment remains uncertain. The independent role of HIV infection on oxidative stress markers during TB treatment remains to be determined. As we did not directly ascertain the predictive value of elevated urine oxidative stress markers with subsequent drug-induced liver injury, prospective studies will be essential to unravel the link between isoniazid and metabolite exposures, hepatocellular oxidative stress, and clinical hepatoxicity events, based on the rigorous assessment of causality with tools such as the Roussel Uclaf Causality Assessment Method (RUCAM) (Danan and Teschke, 2018).
This study had several strengths. Unlike many earlier studies, we examined physiologically relevant isoniazid and metabolite concentrations, supported by a parent-metabolite clinical pharmacokinetic study and physiologically-based pharmacokinetic modelling (Sager et al., 2015). We performed in vitro experiments to provide a mechanistic link between physiologic exposures to these compounds and cellular signals for redox imbalance and oxidative DNA damage. In this approach, we identified a marker of oxidative DNA damage, 8-OHdG, which was strongly induced by hydrazine exposures in the cell model, which we then related to isoniazid pharmacokinetic exposures and GSTA2 genotype in a clinical study. The identification of a urine-based biomarker of oxidative stress related to pharmacokinetic exposures and/or patient genotype supports non-invasive sampling strategies in future prospective studies of isoniazid toxicity. The feasibility of this future work is enhanced by the potential for point-of-care detection of 8-OHdG in paper-based devices (Martins et al., 2017).
In summary, the hydrazine metabolite of isoniazid was the strongest inducer of redox imbalance and downstream oxidative DNA damage in HepG2 cells over long-term exposures to clinically relevant concentrations. In a study of isoniazidtreated TB patients, peak serum isoniazid concentrations and GSTA2 genotype were independently associated with urine levels of 8-OHdG, a marker of oxidative DNA damage. Future clinical studies should evaluate urine 8-OHdG as an early predictor of clinical hepatotoxicity during isoniazid treatment.

DATA AVAILABILITY STATEMENT
The data presented in this study are available in an open data repository (Figshare, doi:10.6084/m9.figshare.12709550).

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the University of Pennsylvania IRB and the Rutgers University IRB. The patients/participants provided their written informed consent to participate in this study.

ACKNOWLEDGMENTS
This manuscript has been released as a pre-print at medRxiv [Zentner et al. (2020)].