Population Pharmacokinetic Analysis of Isoniazid, Acetylisoniazid, and Isonicotinic Acid in Healthy Volunteers

Abstract

In this study, we aimed to quantify the effects of the N-acetyltransferase 2 (NAT2) phenotype on isoniazid (INH) metabolism in vivo and identify other sources of pharmacokinetic variability following single-dose administration in healthy Asian adults. The concentrations of INH and its metabolites acetylisoniazid (AcINH) and isonicotinic acid (INA) in plasma were evaluated in 33 healthy Asians who were also given efavirenz and rifampin. The pharmacokinetics of INH, AcINH, and INA were analyzed using nonlinear mixed-effects modeling (NONMEM) to estimate the population pharmacokinetic parameters and evaluate the relationships between the parameters and the elimination status (fast, intermediate, and slow acetylators), demographic status, and measures of renal and hepatic function. A two-compartment model with first-order absorption best described the INH pharmacokinetics. AcINH and INA data were best described by a two- and a one-compartment model, respectively, linked to the INH model. In the final model for INH, the derived metabolic phenotypes for NAT2 were identified as a significant covariate in the INH clearance, reducing its interindividual variability from 86% to 14%. The INH clearance in fast eliminators was 1.9- and 7.7-fold higher than in intermediate and slow eliminators, respectively (65 versus 35 and 8 liters/h). Creatinine clearance was confirmed as a significant covariate for AcINH clearance. Simulations suggested that the current dosing guidelines (200 mg for 30 to 45 kg and 300 mg for >45 kg) may be suboptimal (3 mg/liter ≤ C_max ≤ 6 mg/liter) irrespective of the acetylator class. The analysis established a model that adequately characterizes INH, AcINH, and INA pharmacokinetics in healthy Asians. Our results refine the NAT2 phenotype-based predictions of the pharmacokinetics for INH.

INTRODUCTION

Due to its high degree of effectiveness and affordability, isoniazid (INH) is a widely used antimycobacterial agent for the first-line therapy of tuberculosis. However, INH also frequently results in liver toxicity and peripheral neuropathy (1, 2), and its metabolism and bioactivation are thought to be critical in INH-induced hepatotoxicity. INH is primarily metabolized through hepatic arylamine N-acetyltransferase type 2 (NAT2) to acetylisoniazid (AcINH) (Fig. 1). INH is also converted to hydrazine and isonicotinic acid (INA) by hydrazinolysis, with the INH to INA biotransformation believed also to be catalyzed by cytochrome P450 (CYP) enzymes, e.g., CYP2C (3). AcINH can be further metabolized to produce acetylhydrazine, which has been proposed as the cause of INH hepatotoxicity via a CYP2E1-mediated bioactivation pathway (4). Since not only the clinical benefit but also the toxicity of INH may be affected by its plasma exposure, the identification of clinical factors that contribute to the variability in INH pharmacokinetics is critical.

FIG 1.

Schematic representation of isoniazid metabolism. Isoniazid is mainly converted by N-acetyltransferase 2 (NAT2) and isoniazid hydrolase into acetylisoniazid and isonicotinic acid and hydrazine, respectively. Then, acetylisoniazid is converted by acetylisoniazid hydrolase into acetylhydrazine and isonicotinic acid, and isonicotinic acid is converted into isonicotinyl glycine. Acetylhydrazine is converted by acyl amidase into monoacetylhydrazine, which is in turn converted by NAT2 into diacetylhydrazine. In this study, the pharmacokinetic data for isoniazid, acetylisoniazid and isonicotinic acid are available for population modeling.

The NAT2 gene, located on chromosome 8p22, demonstrates large interindividual variability in acetylating activities by genetic polymorphisms in humans. At present, more than 40 variants have been identified. INH concentrations are highly variable between subjects, mainly as a result of the variability in activities of these enzymes (5). INH concentrations are associated with antimycobacterial treatment and adverse reactions, and a concentration target of 3 to 6 mg/liter has been suggested from expert opinions, based on the normal values which are usually obtained (6).

The genetic polymorphisms associated with NAT2 activity may result in altered clinical efficacy and/or toxicity of INH (7, 8). As a consequence, dose optimization of INH based on enzyme activity may be warranted. Some have suggested guiding INH dosing based on the NAT2 genotype to prevent subtherapeutic INH concentrations (8). However, NAT2 genotype-based dosing does not account for all nongenetic factors, including NAT2 drug interactions and CYP2E1 activity. Hence, phenotype-guided dosing is likely a more successful dosing strategy.

A population pharmacokinetic approach is often used to develop a model to describe the pharmacokinetics of a drug and to investigate the potential covariates that may contribute to the pharmacokinetic variability. Early population pharmacokinetic studies conducted with healthy North American subjects revealed two subgroups of INH metabolizers (fast and slow) on the basis of differences in clearance values, but interindividual variability in the pharmacokinetic parameters was not reported (9). Similarly, in the absence of NAT2 genotypic or phenotypic data, Wilkins et al. used a mixture model to characterize the dual rates of INH elimination among South African tuberculosis patients (10). Although Kinzig-Schippers et al. identified INH clearance values for fast, intermediate, and slow acetylators, their study was conducted on a small group of Caucasian subjects that included only two intermediate metabolizers (11). It is our understanding that the combined effects of the NAT2-derived phenotype and clinical factors on INH pharmacokinetics have not yet been explored in the Asian population (12). Furthermore, previous studies have described the pharmacokinetics of isoniazid and its metabolites AcINH and INA, but no population pharmacokinetic model for INH and these metabolites is thus far available (13, 14).

Therefore, the aims of this population pharmacokinetic analysis were to develop a model that adequately describes the pharmacokinetics of INH, AcINH, and INA following single oral dose administration in healthy Asian adults and to evaluate the potential impact of the NAT2 metabolic phenotypes, demographics, and measures of renal and hepatic function on INH, AcINH, and INA pharmacokinetics. The analysis was conducted using data from healthy volunteers only because these permit the cleanest assessment of drug disposition without confounding factors often seen in patient data, such as compromised organ function, underlying disease, and/or concomitant medication for comorbidity that may affect INH pharmacokinetics.

MATERIALS AND METHODS

Study subjects.

This was a prospective, open-label, crossover study in healthy adults, with prior stratification by the cytochrome P450 2B6 (CYP2B6) genotype as part of a study examining efavirenz pharmacokinetics in the presence of rifampin (32). A total of 830 people from 2 healthy volunteer cohorts were screened to identify those carrying CYP2B6 516 GG and TT, and 23 GG and 10 TT subjects were recruited. Clinical examinations combined with liver and renal function tests were conducted to verify that the participants were healthy. The subjects abstained from medications, including herbal preparations, 2 weeks before and throughout the study period. All participants gave informed written consent prior to undergoing genotyping and pharmacokinetic evaluation. The study was reviewed and approved by the Institutional Review Board (IRB) of the National Healthcare Group, Singapore.

INH dosing and sampling.

In the main study, the subjects were given a single dose of efavirenz (600 mg) followed by a washout period of 14 days. The subjects were then randomized 1:1 to either rifampin (RIF) (600 mg) or RIF (600 mg)-INH (300 mg) daily for 14 days, followed by another single dose of efavirenz (600 mg), which was administered at the same time as the final dose of RIF or RIF-INH, given in the fasting state (treatment 1). After another washout period, subjects were switched over to RIF (600 mg)-INH (300 mg) or RIF (600 mg) daily, followed by another efavirenz single dose, administered at the same time as the final dose of RIF-INH or RIF, given in the fasting state (treatment 2). For this study, blood samples were drawn at 0, 1, 2, 4, 6, 8, 10, 12, 18, and 24 h after the final dose of INH in the INH treatment arm. Subjects returned to the study site every 2 days for observed dosing and for collection of their RIF or RIF-INH medications for the next day. Compliance with the RIF-INH dosing was further assessed by pill counts and medication diaries.

High-performance LC-MS/MS analysis.

The concentrations of INH, AcINH, and INA in the plasma were measured by a liquid chromatography-tandem mass spectrometry (LC-MS/MS) method validated in our laboratory (15). Briefly, 20-μl plasma samples were diluted with 80 μl of 0.1% (vol/vol) formic acid followed by extraction in 300 μl of methanol containing the internal standards (INH-d₄, AcINH-d₄, and INA-d₄) using Captiva ND^Lipids (Agilent Technologies, Waldbronn, Germany) solid-phase extraction. Serial dilutions were used to obtain linear calibration ranges for INH (5 to 10,000 ng/ml), AcINH (12.5 to 5,000 ng/ml), and INA (12.5 to 5,000 ng/ml). The LC-MS/MS system consisted of an Agilent 1290 binary pump connected to an Agilent 6460 triple quadrupole mass spectrometer (Agilent Technologies). Chromatographic separations were achieved on a Zorbax Aq-SB high-performance liquid chromatography (HPLC) column (Agilent Technologies) with gradient elution. The mass spectrometer was operated under the positive ionization mode, and the detection of INH, AcINH, and INA was based on the multiple reaction monitoring of m/z 138.1 → 121.1, 180 → 121, and 124.1 → 79.9, respectively. The method has been validated according to the FDA guidance for accuracy (89.7 to 113.5%) and precision (relative standard deviation of <11.9%) for each of the high, medium, and low quality controls. The stability of the analytes has also been assessed (mean recovered concentration of 85 to 112%) after 48 h of storage in an autosampler at 6°C and three freeze-thaw cycles. The lower limits of quantification for INH, AcINH, and INA were 5, 12.5, and 12.5 ng/ml, respectively.

NAT2 genotyping and phenotyping.

Genomic DNA was extracted from the blood samples by using an Omega E.Z.N.A. blood DNA minikit (Omega Bio-Tek, Inc., Norcross, GA, USA). Genotyping was conducted by using a Sequenom iPLEX ADME PGx panel v1.0 (Sequenom, Inc., CA, USA). PCR amplifications of the target regions of interest on the iPLEX ADME PGx panel and iPLEX Gold extension chemistry were conducted according to the user's guidelines. Each multiplex was PCR amplified using standard PCR kit conditions against 44 human HapMap control DNAs with known genotypes. After data acquisition on the MassARRAY analyzer, haplotype and copy number variation (CNV) reports were generated by TyperAnalyzer software. NAT2 single nucleotide polymorphisms (SNPs) available on this panel include rs1805158 (c.190 C>T), rs1801279 (c.191 G>A), rs1041983 (c.282 C>T), rs1801280 (c.341 T>C), rs1799929 (c.481 C>T), rs1799930 (c.590 G>A), rs1799931 (c.857 G>A), and rs1208 (c.803 G>A). Classification of the acetylators into poor, intermediate, and fast phenotypes was based on a published algorithm (16).

Population pharmacokinetic analysis.

Population pharmacokinetic models were built using nonlinear, mixed-effects modeling via NONMEM software v7.3 (Icon Development Solutions, Dublin, Ireland) in conjunction with Perl-speaks-NONMEM (PsN) 3.5.3 (17). Compartmental pharmacokinetic models were coded using various ADVAN subroutines in NONMEM. Data exploration, manipulation, and graphics were handled using Xpose 4.3.5 (18) embedded in R 3.1.0 (http://cran.r-project.org/, open-source, S-based statistical software). The first-order conditional estimation method, with eta-epsilon interaction between the inter- and intraindividual variability (FOCE INTER), was used to estimate the pharmacokinetic parameters. Models were fitted to the log-transformed data, and shrinkage on both eta and epsilon was reported. The precision of the parameter estimates was estimated using the covariance step in NONMEM.

Visual inspections of the plasma concentration versus time profiles for INH, AcINH, and INA and the objective function value (OFV) calculated using likelihood ratio tests (P = 0.05) were used to investigate the base model. Various pharmacokinetic models, including one, two, and three compartments with first- or zero-order elimination and saturable and time-varying clearance, were fitted to the data during model development (19–21). The fraction of INH clearance for the formation of AcINH (F_AcINH) and the fraction of AcINH clearance for the formation of INA (F_INA) were estimated by fixing the respective central volumes of distribution for AcINH and INA to 17 liters, which was the calculated volume of distribution of the central compartment of AcINH in Boxenbaum and Riegelman (14). The fraction of INH clearance for the formation of INA was calculated as 1 − F_AcINH. An exponential error model was used to describe the interindividual variability for the pharmacokinetic parameters. The residual variability was modeled as an additive error model. Additive error on log-transformed data corresponds to a proportional or exponential error on a normal scale. Interindividual variability was tested on all pharmacokinetic parameters for which estimation of variability was supported by the data. Consideration was given to fitting a full-block omega structure on the base model, followed by inspection of the correlations among the interindividual variabilities to guide the development of a parsimonious omega structure. Allometric scaling of clearance and volume parameters was tested using the median body weight of 63 kg (22). For example, allometric scaling for CL/F and V₂/F was applied using

$$(\text{CL}/F{)}_i=(\text{CL}/F)\times (\frac{{\text{WT}}_i}{\text{WT}}{)}^{0.75}$$

$$(V_2/F{)}_i=(V_2/F)\times (\frac{{\text{WT}}_i}{\text{WT}}{)}^1$$

where (CL/F₁) is the scaled apparent oral clearance for individual i, CL/F is the typical value of the INH apparent clearance term for a 63-kg individual, and WT_i is the body weight of individual i in kilograms. Similarly, (V₂/F)_i and (V₂/F) are the scaled and typical apparent central volumes of distribution for INH in individual i and in a 63-kg individual, respectively.

Following development of the base model, a total of 12 covariates of clinical interest were systemically evaluated in a stepwise forward selection, and significant covariates were combined in a full model (Table 1). This was followed by backward elimination, and significant covariates were retained in the final model. The continuous covariates tested were age, body surface area (BSA), body mass index (BMI), creatinine, creatinine clearance (estimated using the Cockcroft and Gault equation with total body weight [23]), bilirubin, alkaline phosphatase (ALP), alanine aminotransferase (ALT), and aspartate aminotransferase (AST). The categorical covariates evaluated were gender, race, and NAT2-inferred acetylator status. The covariate selection for the final model was guided using likelihood ratio tests at the following significance levels: P = 0.05 for forward addition of covariates (ΔOFV of 3.84) and P = 0.001 for backward elimination of covariates (ΔOFV of 10.83). In each case, diagnostic plots and comparisons of changes in the minimum OFV between the nested models were used to evaluate the covariates. The effects of a continuous covariate on a parameter were represented as a power function referenced to the median value in the population: TVP_i = θ₁ · (COV_i/COV_ref)^θ2, where TVP_i is the typical value of a pharmacokinetic parameter (P) for an individual i with a COV_i value of the covariate, θ₁ is the typical value for an individual with a reference covariate value of COV_ref, and θ₂ is the exponent of the power function. The effects of a categorical covariate on a parameter were represented through TVP_i = θ₁ · (1 + θ₂ · IND_i), where θ₁ is the typical parameter value for an individual in the absence of the covariate (IND_i = 0) and θ₂ is the fractional change in the typical value for an individual if the covariate is present (IND_i ≠ 0).

TABLE 1.

Main covariates in the population pharmacokinetic analysis of isoniazid, acetylisoniazid, and isonicotinic acid (n = 33)

Parameter	Value
Continuous covariates (median [range])
Age (yr)	33 (22–56)
Body weight (kg)	62.5 (45.8–86.1)
Height (cm)	168 (146.8–179)
Body surface area (m2)	1.71 (1.37–1.99)
Body mass index (kg · m−2)	22.9 (17.5–30.5)
Serum creatinine (μmol · liter−1)	70 (43–101)
Creatinine clearance (ml · min−1)	113.1 (52.3–174.6)
Total bilirubin (μmol · liter−1)	8.8 (3–20)
ALP (U · liter−1)	60 (28–90)
ALT (U · liter−1)	18 (7–64)
AST (U · liter−1)	20 (14–61)
Categorical covariates (no. [%])
Gender (male/female)	23 (70)/10 (30)
Race (Chinese/Malay/Indian)	21 (64)/7 (21)/5 (15)
Acetylator (fast/intermediate/slow)	7 (21)/15 (45)/11 (33)

During the development of the model, all 12 covariates were tested on the clearance parameters. Body weight, BMI, BSA, age, race, gender, and NAT2 acetylator status were tested on the volume parameters. The final model selection was based on the evaluation of goodness-of-fit plots, the successful convergence and plausibility of the parameter estimates, and the OFVs calculated using likelihood ratio tests. The shrinkage percentage was calculated for each interindividual variability and for the residual error (24). The adequacy of the final model was assessed by conducting a visual predictive check (VPC) (n = 1,000) between the observed and simulated data (25).

The final population pharmacokinetic model was used to simulate the INH concentration versus time profiles (n = 1,000) and exposure after administration of a single 200- or 300-mg dose in fast, intermediate, and slow acetylators. These doses reflect the WHO daily dose guidelines for subjects belonging to the 30- to 45-kg and >45-kg weight bands, respectively (26). The maximum plasma INH concentration (C_max) was derived for each simulated subject. The distribution of C_max was compared with the previously published cutoffs of 3 mg/liter (21.9 μmol/liter) and 6 mg/liter (43.8 μmol/liter) (6).

RESULTS

Totals of 298, 309, and 310 INH, AcINH, and INA concentrations, respectively, were available for population pharmacokinetic modeling. Overall, data for 7 fast, 15 intermediate, and 11 slow metabolizers, respectively, were available (Table 1). The base model used to describe INH pharmacokinetics was a linear two-compartment structural model. The pharmacokinetics of AcINH and INA were best described using linear two- and one-compartment model, respectively, linked to the INH model (Fig. 2). The base model was parameterized in terms of the first-order rate constant of INH absorption (k_a), apparent INH clearance (CL/F), apparent central volume of distribution for INH (V₂/F), apparent intercompartmental clearance for INH (Q/F), apparent peripheral volume of distribution for INH (V₅/F), fraction of INH clearance for the formation of AcINH (F_AcINH), AcINH clearance (CL_A), central volume of distribution for AcINH (V₃), intercompartmental clearance for AcINH (Q_A), peripheral volume of distribution for AcINH (V₆), fraction of AcINH clearance for the formation of INA (F_INA), INA clearance (CL_I), and volume of distribution for INA (V₄) (Fig. 1). Random effects characterizing interindividual variability were added on k_a, CL/F, V₅/F, CL_A, V₆, CL_I, and oral bioavailability of INH (F_INH). The effect of body weight on clearance and volume parameters was included in the base model via allometric scaling (22). Competing models in which BSA and BMI were included as covariates for clearance and volume parameters were not comparatively superior, with larger OFVs and smaller contributions to the interindividual differences in these parameters. Moreover, with the allometric body weight relationships, no clear relationships were discerned between BSA or BMI and volume/allometric body weight or clearance/allometric body weight. Inclusion of allometric scaling on the clearance and volume terms resulted in a significant decrease in the NONMEM OFV. In addition, the goodness-of-fit plots showed less bias, and interindividual variability in most clearance and volume terms were lower relative to those generated from the model without allometric scaling.

FIG 2.

Schematic representation of the population pharmacokinetic model of isoniazid (INH) and its metabolites acetylisoniazid (AcINH) and isonicotinic acid (INA). The absorption of INH is described with a rate constant (k_a). The oral bioavailability of INH (F_INH) is not estimated due to a lack of pharmacokinetic data from intravenous dosing. Drug transfer between the central and peripheral INH compartments is described with the intercompartmental clearance Q/F. The equilibrium of the central AcINH compartment with its peripheral compartment is described with the intercompartmental clearance parameter Q_A. The metabolism of INH to AcINH and INA is described with the clearance parameters F_AcINH · (CL/F) and (1 − F_AcINH) · (CL/F), respectively. Metabolism of AcINH to INA is described with the clearance parameter F_INA · CL_A. The clearance of INA is described with the clearance parameter CL_I. From this model, the volumes of distribution for central (V₂/F) and peripheral (V₅/F) INH and peripheral AcINH (V₆) can be estimated. The volumes of distribution for central AcINH and INA are fixed to a volume of 17 liters to ensure model identifiability.

Estimated INH, AcINH, and INA pharmacokinetic parameters from the base model and final model are shown in Table 2, and the goodness-of-fit plots for the final model are shown in Fig. 3. In the final model, the relative standard error obtained from the covariance step was <25% for all fixed-effect parameters. Two additional covariate relationships were assessed to be significant at the P = 0.001 level. The CL/F values in fast eliminators were 1.9- and 7.7-fold higher than in intermediate and slow eliminators (65 versus 35 and 8 liters/h). Creatinine clearance was confirmed as a significant covariate for AcINH clearance, reducing the interindividual variability of CL_A from 15.4% in the base model to 10.3% in the final model. Addition of the NAT2 metabolic phenotype as a covariate on CL/F decreased the interindividual variability from 86.1% in the base model to 14.4% in the final model. Shrinkage for all etas was <31%, which was considered adequate for use of empirical Bayes estimates for diagnostic plots. Shrinkage on the residual error models for INH, AcINH, and INA was <11%. As shown in Fig. 3, data (population and individual predictions versus observations) were evenly and randomly distributed across the line of identity, indicating that there was no major bias and that the final model is appropriate for the population and each individual. An examination of the final model goodness-of-fit plots revealed that conditional weighted residuals were evenly distributed about zero (data not shown), indicating no major bias in the structural and residual error models.

TABLE 2.

Estimated pharmacokinetic parameters in the base model and final model

Parametera	Base model estimates			Final model estimates
	Mean (% RSEb)	% IIVc (% RSE)	% shrinkage	Mean (% RSE)	% IIV (% RSE)	% shrinkage
ka (1/h)	0.6 (6.3)	12.6 (50.9)	40	0.6 (5.3)	12.9 (43.6)	31
CL/F (liters/h)	25.1 (14.5)	86.1 (21.3)	31.6	65.2 (6.9)	14.4 (47.9)	24
V2/F (liter)	16.2 (35.2)			18 (22.6)
Q/F (liters/h)	2.9 (16.6)			2.8 (14.5)
V5/F (liters)	16.5 (12.8)	34.9 (38)	15	15.9 (11.3)	34.4 (36)	13.6
FINH	1	38.1 (32.3)	2.4	1	31.6 (28.3)	4.3
FAcINH	0.965 (1.7)			0.973 (1.3)
CLA (liters/h)	21.3 (7.8)	15.4 (60.9)	20.2	21.3 (7.2)	10.3 (30.1)	7.6
V3 (liters)	17			17
QA (liters/h)	71.5 (20)			69.2 (13.2)
V6 (liters)	81.2 (10.5)	19.4 (54.8)	19.4	80.4 (9.4)	20.4 (34.4)	9.1
FINA	0.7 (11.2)			0.734 (10.1)
CLI (liters/h)	45.1 (5.9)	19.5 (42.5)	17.3	44.6 (5.6)	18.6 (46.5)	20.8
V4 (liters)	17			17
Intermediate acetylator on CL/Fd				0.5 (9.1)
Slow acetylator on CL/Fd				0.9 (1.8)
CrCL on CLAe				0.4 (21.5)
Residual errorf
INH [log(mg/liter)]	0.326 (15.1)		11.5	0.326 (14.3)		10.6
AcINH [log(mg/liter)]	0.207 (23.5)		9.6	0.207 (24.1)		9.6
INA [log(mg/liter)]	0.270 (14.5)		9.2	0.269 (13.8)		7.3

^ak_a, first-order rate constant of INH absorption; CL/F, apparent INH clearance; V₂/F, apparent central volume of distribution for INH; Q/F, apparent intercompartmental clearance for INH; V₅/F, apparent peripheral volume of distribution for INH; F_INH, oral bioavailability for INH; F_AcINH, fraction of INH clearance for the formation of AcINH; CL_A, AcINH clearance; V₃, central volume of distribution for AcINH; Q_A, intercompartmental clearance for AcINH; V₆, peripheral volume of distribution for AcINH; F_INA, fraction of AcINH clearance for the formation of INA; CL_I, INA clearance; V₄, volume of distribution for INA.

^bRSE, relative standard error.

^c% IIV, % coefficient of variance.

^dAdding NAT2 phenotype as a covariate on CL/F decreased IIV from 86.1% in the base model to 14.4% in the final model.

^eAdding creatinine clearance (CL_CR) as a covariate on CL_A decreased IIV from 15.4% in the base model to 10.3% in the final model.

^fAdditive error model in log scale.

FIG 3.

Goodness-of-fit plots of the final model for isoniazid (INH) (left panels), acetylisoniazid (AcINH) (middle panels), and isonicotinic acid (INA) (right panels). The top panels show the observed concentrations versus the population-predicted concentrations; the bottom panels show the observed concentrations versus the individual predicted concentrations. The solid line in all subplots represents the line of identity. The dashed line in all subplots represents the loess smoothing.

The predictive performance of the final model was assessed by VPC. A total of 1,000 replicates of the data set were simulated using the final population pharmacokinetic model following single oral dosing with INH (300 mg). As shown in Fig. 4, the observed medians and 5th and 95th data percentiles were adequately captured by the corresponding simulation-based 95% confidence intervals (shaded areas) around the predicted medians and 90th percentiles for INH, AcINH, and INA.

FIG 4.

Visual predictive check for the final isoniazid (top panel), acetylisoniazid (middle panel), and isonicotinic acid (bottom panel) population pharmacokinetic model. In each subplot, the solid line connects the observed median values, whereas the dashed lines represent the observed 5th and 95th percentiles of the observations. The light gray areas indicate the 95% confidence intervals (CIs) of the 5th and 95th percentiles of the predicted values, whereas the dark gray areas indicate the CIs of the medians.

Figure 5 compares the box plots of predicted INH C_max values obtained from single 200- and 300-mg doses in fast, intermediate, and slow acetylators. For the 200-mg dose scenario, 39.6 and 7.4% of the simulated fast and intermediate acetylators, respectively, had a C_max of <3 mg/liter (or, equivalently, 21.9 μmol/liter). In addition, 10.4, 42.4, and 97.3% of the simulated C_max values for fast, intermediate, and slow acetylators, respectively, were >6 mg/liter (43.8 μmol/liter). Following administration of 300 mg INH, 37.6 and 6.7% of the simulated C_max values for fast and intermediate acetylators, respectively, were <3 mg/liter. The data further showed that 9.6, 43.3, and 98.3% of the simulated fast, intermediate, and slow acetylators, respectively, had C_max values of >6 mg/liter. Overall, at the current WHO dosing guidelines, fast acetylators may be at risk for subtherapeutic efficacy, whereas intermediate and slow acetylators may need lower doses to achieve optimal INH plasma exposure.

FIG 5.

Box plots of simulated maximum isoniazid (INH) concentrations after a single 200-mg (left panel) or 300-mg (right panel) dose in fast, intermediate, and slow acetylators. The 200- and 300-mg INH doses were administered to subjects belonging to the 30- to 45-kg and >45-kg weight bands, respectively. For each subplot, the lower and upper dashed lines denote INH concentrations of 3 mg/liter (21.9 μmol/liter) and 6 mg/liter (43.8 μmol/liter), respectively. For each box plot, the thick line is the median, the box represents the interquartile range (IQR), the whiskers extend to the most extreme data point that is <1.5 times the IQR, and the open circles are outliers.

DISCUSSION

To our knowledge, this is the first report on the population pharmacokinetic parameters of INH and its two major metabolites, AcINH and INA. The INH population pharmacokinetic parameters in a healthy Asian population were well described by a two-compartment, mixed-effects model with a first-order absorption and linear elimination process. These findings are consistent with previously reported results from compartmental analysis of INH pharmacokinetic data (10, 11). The population pharmacokinetics of AcINH and INA were described by a two- and one-metabolite compartment model, respectively, which are also in line with the structural models of the respective metabolites in previous studies (13, 14). Due to the unavailability of data for urinary excretion of INH and its metabolites and the unknown fraction of parent drug that is converted to metabolite, the model was kept identifiable by fixing the central volumes of distribution of AcINH and INA to a previously found central volume of distribution for AcINH (14).

All clearance and volume terms in our population pharmacokinetic model were allometrically scaled by body weight, as it improved and stabilized the model prior to covariate testing. Creatinine clearance (CL_CR) was a statistically significant covariate for the clearance of AcINH but not INH and INA. For every 10% increase or decrease in CL_CR from 113 ml/min, the change in CL_A/F was marginal (within ±5%), and therefore, an INH dose adjustment was deemed unnecessary based on renal function according to the observation in reference 6. None of the other biological covariates tested had any clinically meaningful influence on INH, AcINH, and INA pharmacokinetics. The absence of a significant association between the pharmacokinetic parameters and bilirubin, ALP, ALT, or AST in healthy volunteers was not unexpected, since the levels of these laboratory measurements were within the normal ranges and indicative of normal kidney and liver function. The lack of an effect of gender on the INH central volume of distribution observed in the healthy volunteers in this analysis is similar to the finding reported for healthy North Americans (11), but not to that of Wilkins et al. (10), who reported a lower INH central volume of distribution in female than in male tuberculosis patients. A possible explanation for this discrepancy is a lack of difference in the percentage of body fat between the female and male subjects in our study (27).

In line with the existing data implicating NAT2 as the major pathway of clearance for INH, our analysis showed that NAT2 metabolic status contributes to a substantial 72% of INH pharmacokinetic variability in the subjects investigated. Our study quantified, for the first time, the average CL/F values in fast (n = 7), intermediate (n = 15), and slow (n = 11) INH eliminators among Asian adults. Slow eliminators may lack a functional NAT2 enzyme, and hence, the CL/F in these subjects could represent INH clearance not attributed to NAT2 (10). Furthermore, given that INH is predominantly metabolized in the liver via acetylation, it is likely that the higher CL/F in fast eliminators stems mainly from increased clearance and not from reduced bioavailability. This also explains why the volume of distribution of INH is not different among fast, intermediate, and slow acetylators. The magnitude of CL/F in the slow acetylators in our study (8.5 liters/h) was in agreement with values observed in recent studies involving Caucasians and South Africans (4.7 to 15 liters/h) (9–11). However, CL/F in the fast acetylators in our study (65.2 liters/h) was approximately 1.3- to 5-fold larger relative to that in these studies. These results indicate that, apart from NAT2 polymorphism, geographic and ethnic differences may have a profound effect on INH clearance. Another possible cause is differences between the cotreatments administered to subjects in different studies.

Dosing simulations via the final model were performed to calculate the population proportion achieving pharmacodynamically relevant C_max values (3 mg/liter ≤ C_max ≤ 6 mg/liter) following single-dose INH treatment (200 or 300 mg for 30 to 45 kg and >45 kg of body weight, respectively). These targets are based on concentrations that are usually achieved in patients who undergo successful treatment and are above the MIC of drug-susceptible Mycobacterium tuberculosis (≤0.4 mg/liter).

These simulations were necessary, as the different INH clearance values in the fast, intermediate, and slow acetylators result in considerably different plasma INH exposures across these three groups of subjects. The simulations indicate that as many as 38% of the fast acetylators may exhibit C_max values below the lower cutoff limit of 3 mg/liter, and a higher dose may be required in this subgroup. In addition, the current recommended INH dose also may not be optimal for the intermediate and slow acetylators, since a substantial proportion (42 to 98%) of these subjects might experience supratherapeutic concentration levels (C_max of >6 mg/liter). For these subjects, a lower INH dose may be warranted. Overall, our results indicate that the current dosing recommendations for INH may not result in optimal INH exposure in subjects with different metabolic phenotypes for NAT2.

Although the Monte Carlo simulation analysis appropriately handles the issues of between-subject variability and covariate inclusion, several potential limitations must be considered. First, a relatively small sample size (33 subjects) was used in the current study, and this may not represent the true pharmacokinetic variability in the population, especially that of tuberculosis patients. However, the recruitment of large numbers of patients is challenging, and simulations based on small cohorts are instructive for assessing various dosing strategies, especially when it is clear that a large change in dosage from the usual is required to attain suitable concentrations. A second limitation is the use of published C_max data for comparison against the simulated concentration values. Although a range of C_max values was used in this study, the relevancy and applicability of this pharmacodynamic measure may be inadequate because susceptibility to Mycobacterium tuberculosis may vary over time, between countries, and between patient profiles. On this basis, further studies to elucidate exposure-response relationships are warranted to assess the significance of our simulation results and to optimize INH dosing in the fast, intermediate, and slow acetylators. A third limitation is that this was not a study dedicated to studying isoniazid alone. However, we believe that prior CYP2B6 stratification and the coadministration of a single dose of efavirenz were unlikely to affect the isoniazid pharmacokinetics.

The pharmacokinetic variability that is driven largely by NAT2 creates an opportunity and a challenge for the deployment of INH in patients. Therapeutic drug monitoring has therefore been suggested to achieve targeted C_max values, especially for cases of slow response or toxicity (6). Our simulations suggest that tailoring the INH dosing to NAT2 metabolizer status may be necessary and is expected to result in an improved balance between risk and benefit upon treatment. NAT2 is a highly polymorphic gene, with new discoveries arising on a regular basis. More than 40 allelic variants have been discovered to date, resulting in a broad range of functional variations. The NAT2 metabolizer status can also be measured using the acetylisoniazid/isoniazid metabolic ratio in plasma (15), but this test is not widely available.

This complex variability is unlikely to be adequately described with a three-bin categorization of the enzymatic capacity system. It is inevitable that a range of functions will remain within some or all of such categories, perhaps with the exception of the category associated with a complete loss of activity (slow eliminator). We have noticed that the interindividual variability in CL/F cannot be estimated separately for the fast, intermediate, and slow eliminator subgroups. Nonetheless, the relatively small sample size in our study is an important limitation that can only be addressed with further clinical work.

Notable characteristics of our population model were an interindividual variability of 31.6% for F_INH and intrasubject variabilities (additive errors in log scale) of 0.326, 0.207, and 0.269 mg/liter for INH, AcINH, and INA, respectively. The exact cause(s) of the interindividual variability in oral bioavailability of INH has yet to be determined. However, because CYP2E1, CYP2C, or CYP3A4 has been proposed to play a key role in the metabolism of INH, it is speculated that a major source of variability may be differences in the CYP450 expression and/or activity in the intestine and liver (28, 29). Indeed, 10- to 40-fold variabilities in the expression of CYP3A4 have been reported in healthy subjects (30). Further studies are suggested to determine whether and how CYP450, including CYP3A, is involved in INH metabolism. With regard to the comparatively higher residual variability in INH pharmacokinetic data, it is plausible that differences in the dissolution and ensuing gastrointestinal absorption of INH may be a contributing factor. Since the solubility of INH is pH dependent (31), alterations in pH values in the stomach and duodenum may lead to variable dissolution of INH.

In conclusion, we developed a population pharmacokinetic model for INH and its metabolites AcINH and INA that incorporates NAT2 metabolic phenotypes. These metabolic phenotypes explained a significant part of the variability in INH clearance. In addition, creatinine clearance had a major impact on the clearance of AcINH. Our finding concerning suboptimal plasma INH exposure based on current WHO dosing guidelines warrants further investigation and confirmation in larger populations, especially those involving tuberculosis patients.