Population Pharmacokinetics and Bayesian Dose Adjustment to Advance TDM of Anti-TB Drugs.

Tuberculosis (TB) is still the number one cause of death due to an infectious disease. Pharmacokinetics and pharmacodynamics of anti-TB drugs are key in the optimization of TB treatment and help to prevent slow response to treatment, acquired drug resistance, and adverse drug effects. The aim of this review was to provide an update on the pharmacokinetics and pharmacodynamics of anti-TB drugs and to show how population pharmacokinetics and Bayesian dose adjustment can be used to optimize treatment. We cover aspects on preclinical, clinical, and population pharmacokinetics of different drugs used for drug-susceptible TB and multidrug-resistant TB. Moreover, we include available data to support therapeutic drug monitoring of these drugs and known pharmacokinetic and pharmacodynamic targets that can be used for optimization of therapy. We have identified a wide range of population pharmacokinetic models for first- and second-line drugs used for TB, which included models built on NONMEM, Pmetrics, ADAPT, MWPharm, Monolix, Phoenix, and NPEM2 software. The first population models were built for isoniazid and rifampicin; however, in recent years, more data have emerged for both new anti-TB drugs, but also for defining targets of older anti-TB drugs. Since the introduction of therapeutic drug monitoring for TB over 3 decades ago, further development of therapeutic drug monitoring in TB next steps will again depend on academic and clinical initiatives. We recommend close collaboration between researchers and the World Health Organization to provide important guideline updates regarding therapeutic drug monitoring and pharmacokinetics/pharmacodynamics.

Key Points

Introduction

Tuberculosis (TB) is still the number one cause of death due to an infectious disease. The World Health Organization estimated that about 1.6 million individuals died from TB in 2018. Of the 10 million people that developed TB in 2018, 3.4% of new cases and 18% of previously treated TB cases were estimated to carry a drug-resistant TB strain [1]. Most of the new cases with drug-resistant TB were diagnosed with multi-drug-resistant (MDR)-TB, defined as TB that is resistant to at least isoniazid and rifampicin. Despite all current efforts, the global burden of disease is not falling fast enough to reach the milestones of the ‘WHO End TB Strategy’ [1]. Next to that, the COVID-19 pandemic will further jeopardize TB control [2]. Providing patients with more active and less toxic treatment may help to achieve goals. In addition to the development of new drugs, optimization of current treatment is important to ensure that drugs for drug-susceptible (DS) and drug-resistant TB are given at doses (and achieve exposures) most likely to be effective without causing significant toxicity. The aim of this narrative review is to summarize the knowledge on the pharmacokinetics and pharmacodynamics of the currently available anti-TB drugs and show how population pharmacokinetics coupled with therapeutic drug monitoring (TDM) can be used to optimize treatment for individual patients.

Pharmacokinetics and Pharmacodynamics of Anti-TB Drugs

Pharmacokinetics describes the behavior of a drug in the patient’s body, including absorption, distribution, metabolism, and excretion, whereas pharmacodynamics describes the biochemical or pharmacological effect of a drug at the site of action in the patient’s body. Over the past decade, pharmacokinetic (PK) and pharmacodynamic (PD) science has greatly contributed to the understanding of TB treatment response and outcome [3]. In infectious diseases, it is common to link exposure of the drug to the sensitivity of the bacteria, the minimum inhibitory concentration (MIC). In TB, the PK/PD indexes are similar to those for other infectious diseases; maximal plasma concentration (Cmax)/MIC, area under the concentration–time curve of a dosing interval (AUC0–τ)/MIC, and time of the concentration over MIC (T > MIC). Often, the total concentration, i.e., the sum of the unbound and plasma protein bound fraction of the drug, is determined. Only the free (f) drug can diffuse through biological membranes to the site of infection and exert its pharmacological effect [4]. Therefore, especially for highly bound drugs, the above indexes should be converted to fCmax/MIC and fAUC0–τ/MIC. For dose optimization, most studies aim at achieving drug exposure associated with efficacy in 80% of the patients. In clinical practice, MIC determinations and free drug concentrations are not measured because of logistical or financial limitations. In such cases, low exposure of a certain drug is often reported as a risk factor for suboptimal treatment. As there is no formal definition of ‘low concentration’, we have defined low exposure to be < 50% inhibitory concentration, which is the concentration that causes 50% of the maximum kill. This is, in our opinion, more meaningful than expressing it as a percentage of exposure achieved in healthy volunteers because that does not include a measure of response.

For most drugs, specific PK/PD indexes have been linked to their efficacy, reflecting their time- or concentration-dependent activity. However, it is important to note that Cmax and AUC0–τ are closely linked and therefore their PK/PD indexes as well.

Hollow fiber infection models have been very useful to explore PK/PD relationships as it is possible to simulate human concentration–time profiles, and test different dosing strategies [5]. Although hollow fiber infection models have limitations, i.e., a lack of human immune response and a maximum number of drugs that can be tested simultaneously, the results correlate well with treatment response [6], which resulted in endorsement by the European Medicines Agency to make use of these models to guide dose selection for clinical trials [7].

Population Pharmacokinetics

The typical PK profile of a drug and the variability among individuals can be characterized with population PK models. Also known as nonlinear mixed-effect models, this is a type of compartmental model consisting of a structural part describing the typical behavior, and a stochastic part accounting for random variability [8]. The structural part includes the primary PK parameters such as the rate of absorption, volume of distribution, and clearance, as well as a quantitative estimate of the influence of patient characteristics (covariates, e.g., body size, age, genetic polymorphisms) on those parameters. The stochastic part describes the magnitude and distribution of random variability between individuals and within an individual between different dosing occasions. It also contains the unexplained variability, i.e., the residual error. Population PK models are developed using drug concentration measurements gathered after controlled dosing. The development and evaluation of the population PK models are done according to widely accepted standards in the field [9], generally employing software packages such as NONMEM (ICON plc, Gaithersburg, MD, USA) and Monolix (Lixoft, Antony, France). An overview of the set-up and development using population PK models is presented in Fig. 1.

Fig. 1

Population pharmacokinetic (PK) modeling. incl. including, MIC minimum inhibitory concentration, PD pharmacodynamic, TDM therapeutic drug monitoring. Created with Biorender.com

TDM and Dose Adjustment

Therapeutic drug monitoring is a tool used to integrate PK and PD knowledge to optimize and personalize drug therapy. Therapeutic drug monitoring uses drug plasma concentrations to personalize drug therapy to bring and keep the concentration within the targeted therapeutic range. Below this range, the treatment is less effective, whereas high concentrations may result in toxicity [4]. When selecting a population PK model for TDM purposes, it is important to consider if the population in which the model was developed is representative for the population where TDM is to be conducted.

Therapeutic drug monitoring is mostly performed using blood samples, i.e., serum, plasma, or dried blood spot, while the TB infection is located elsewhere, in the lungs or extra-pulmonary, for instance, in the central nervous system in the case of TB meningitis. Drug penetration at the site of mycobacterial infection may be hampered, if the disease progresses and caseous lesions are formed [10]. Strydom et al. investigated the penetration of nine anti-TB drugs in resected lung tissue and showed that, irrespective of the drug choice, patients with cavitary disease were at risk for subtherapeutic concentrations [10].

Derived or secondary PK parameters commonly used in TDM such as AUC0–τ, Cmax, and trough concentration (Cmin) can be predicted for specific individuals based on a population PK model, relevant dosing, and covariate information plus any observed drug concentrations. Correct handling of inter-occasion variability is important to maximize the precision of the prediction [11].

For most anti-TB drugs, the AUC0–τ (often AUC0–24 h)/MIC is the most predictive PK/PD index. However, a specific AUC0–24 h/MIC target value has not yet been established for all anti-TB drugs. Here, it must be addressed that individual MIC values might not always be reliable and often are not available early in treatment. Moreover, MIC values might be significantly different when using different methods (e.g., solid or liquid media) [5, 12, 13]. This has important implications for the interpretation of AUC0–24 h/MIC target values when taking into account how the MIC was determined in that study.

Obtaining a full PK curve to calculate the AUC0–24 h using non-compartmental analysis techniques is a laborious and expensive procedure and is thus not feasible in clinical practice. A limited sampling strategy (LSS) either based on a population PK model or on multiple linear regression may help to overcome these problems. This method implies that a limited number, usually one to three, of appropriately timed blood samples is needed to adequately predict the AUC0–24 h as a measure for drug exposure [14]. Population PK models are often used in clinical dosing software that aids in estimating PK parameters such as AUC and that uses Bayesian simulations to estimate appropriate drug exposure [15, 16].

Bayesian-informed dosing takes into account the prior information (including patient-specific values of covariates such as body size, age, and genetics) from a population PK model and uses individual measurements, usually drug concentrations, to provide a patient-tailored posterior dosing recommendation. This allows individual patients plasma drug exposures to meet the PK/PD target [17]. The advantages of the Bayesian approach are the flexible timing of samples, as the population PK model can correct for deviations, and that it takes a number of covariates into account leading to a more accurate estimation of AUC0–24 h [18]. The advantage of multiple linear regression-based LSSs is that these do not require modeling software and AUC0–24 h can be easily estimated using only an equation and the measurement of drug concentrations. The disadvantage is that samples should be taken as close as possible to predefined schedules, and the population of interest should be comparable to the population in which the multiple linear regression has been performed because patient characteristics are not included in the equations to estimate drug exposure [18].

Anti-TB Drugs

In the section below, we summarize the mechanism of action, PK/PD drivers from preclinical models, human pharmacokinetics (with sources of variability, where known), and clinical PK/PD targets for each drug that is used in the treatment of drug-susceptible or MDR-TB. An overview on relevant PK models is included and recommendations on sampling strategies and dosing strategies are summarized. Table 1 gives an overview of the published population PK models per anti-TB drug. Table 2 indicates PK/PD targets and normal ranges of exposures that have been identified for the different drugs, this is not repeated in the text.

Table 1

Overview of population pharmacokinetic models of anti-tuberculosis drugs

First author, year	Software	Covariate–parameter relationships	Population, country of origin (n)	Model structure
Isoniazid
Peloquin, 1997 [25]	NPEM2, IT2B	–	Healthy male adults (24), USA	No model specified, NPEM and IT2B analysis
Wilkins, 2011 [174]	NONMEM	Sex, V HIV, CL Acetylation status, CL Body weight, CL Body weight, V	Patients with TB, adults (235), South Africa	Two-compartment, first-order absorption and elimination with allometric scaling
Zvada, 2014 [175]	NONMEM	NAT2 genotype, CL NAT2 genotype, F Body weight, CL Body weight, V	Patients with TB, children (76), South Africa	Two-compartment, first-order elimination, distribution with absorption transit compartments
Hiruy, 2015 [181]	ADAPT5	–	Patients with TB (children 31), South Africa	Two-compartment
Seng, 2015 [182]	NONMEM	NAT2 genotype, CL Body weight, CL Body weight, V	Healthy adults (33), Singapore	Two-compartment, first-order absorption
Denti, 2015 [76]	NONMEM	Acetylation status, CL Body weight, CL Fat-free mass, V	Patients with TB, adults (100), Tanzania	Two-compartment disposition, transit compartment absorption
Lalande, 2015 [183]	Pmetrics	Acetylation status, Ke	Adults with and without AIDS, volunteers without TB (89)	Three-compartment
Rockwood, 2016 [184]	Monolix	HIV, CL NAT2 genotype, CL	Patients with TB/HIV (100), South Africa	Two-compartment, first-order elimination and absorption through transit compartments
Vinnard, 2017 [185]	NONMEM	NAT2 genotype, CL	Patients with TB (40), Botswana	Two-compartment, first-order elimination
Horita, 2018 [30]	Monolix	NAT2 genotype, CL	Patients with TB, children (113), Ghana	Two-compartment, first-order absorption and linear elimination
Chirehwa, 2019 [186]	NONMEM	Acetylation status, CL Fat-free mass, CL Fat-free mass, V	Patients with TB/HIV, adults (150), Benin and Guinea	Two-compartment disposition, lag time in absorption, liver compartment
Aruldhas, 2019 [187]	NONMEM	Body weight, CL Body weight, V Acetylation status, CL	Patients with TB, children (41), India	One-compartment disposition, absorption phase with transit compartments
Rifampicin
Peloquin, 1997 [25]	NPEM2, IT2B	–	Healthy male adults (24), USA
Wilkins, 2008 [176]	NONMEM	FDC/SDF, CL FDC/SDF, MTT	Patients with pulmonary TB (261), South Africa	One-compartment, transit compartment
Goutelle, 2009 [188]	NONMEM	–	Patients with HIV (20) and healthy volunteers (20), USA	Three-compartment
Smythe, 2012 [189]	NONMEM	Normal fat mass, CL Normal fat mass, V HIV, V	Patients with pulmonary TB (174), Africa	One-compartment, transit compartment, auto-induction
Milán-Segovia, 2013 [190]	NONMEM	Sex, CL Sex, V	Patients with TB (171), Mexico	One-compartment, lag time
Hiruy, 2015 [181]	ADAPT5	–	Patients with TB (children 31), South Africa	One-compartment
Jeremiah, 2014 [191]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with TB/HIV (100), Tanzania	One-compartment, first-order absorption, transit compartment absorption
Seng, 2015 [192]	NONMEM	Body weight, CL Body weight, V	Healthy adults (34), Asia	One-compartment, transit compartment
Sturkenboom, 2015 [14]	MWPharm	–	Patients with TB (55), the Netherlands	One-compartment
Jing, 2015 [193]	NONMEM	–	Patients with pulmonary TB (54), China	One-compartment
Savic, 2015 [15]	NONMEM	Body weight, CL Body weight, V	Patients with TB meningitis (children, 53), Indonesia	Two-compartment
Denti, 2015 [194]	NONMEM	Body weight, CL Body weight, V	Pregnant women with TB/HIV (48), South Africa	One-compartment, first-order elimination, transit compartment absorption
Schipani, 2016 [195]	NONMEM	Age, F Age, CL Body weight, CL Body weight, V	Patients with TB, adults (115) and children (50), Malawi	One-compartment
Chirehwa, 2016 [39]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with TB/HIV (61), South Africa	One-compartment, transit compartment, auto-induction
Rockwood, 2016 [184]	Monolix	HIV, CL lopinavir/ritonavir based antiretroviral regimen, CL	Patients with TB/HIV (100), South Africa	One-compartment, first-order elimination and first-order absorption, with an absorption lag time
Svensson, 2017 [46]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with pulmonary TB (83), South Africa	One-compartment, transit compartment, auto-induction
Horita, 2018 [30]	Monolix	–	Patients with TB, children (113), Ghana	One-compartment, sequential zero- and first-order absorption, first-order elimination
Svensson, 2018 [44]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with TB (336), Tanzania and South Africa	One distribution compartment, absorption through a dynamic transit compartment, Michelis–Menten function limiting the clearance at high concentrations
Svensson, 2019 [38]	NONMEM	Fat-free mass, CL Fat-free mass, V	Adult patients with TB (133), Indonesia	Two disposition compartments, saturable clearance, and autoinduction
Pyrazinamide
Zhu, 2002 [196]	PASTRX, NPEM2, USC*PACK	–	Adult patients with TB (67), children (23), USA	No model specified, NPEM analysis
Hiruy, 2015 [181]	ADAPT5	–	Patients with TB (children 31), South Africa	One-compartment
Denti, 2015 [76]	NONMEM	Fat-free mass, CL Fat-free mass, V NAT2 acetylation status, CL	Patients with TB, adults (100), Tanzania	One-compartment, first-order elimination, One-compartment, first-order elimination, transit compartment absorption
Rockwood, 2016 [184]	Monolix	–	Patients with TB/HIV (100), South Africa	One-compartment, first-order elimination, transit compartment absorption
Vinnard, 2017 [65]	NONMEM	Sex, CL Body weight, CL Body weight, V	Patients with TB/HIV (40), Botswana	One-compartment, first-order elimination
Chirehwa, 2017 [58]	NONMEM	–	Patients with TB/HIV (61), South Africa	One-compartment, first-order elimination, transit compartment absorption
Horita, 2018 [30]	Monolix	–	Patients with TB, children (113), Ghana	One-compartment, with transit compartment absorption and first-order elimination
Mugabo, 2019 [197]	Monolix	–	Patients with MDR-TB (51), South Africa	One-compartment, transit compartment absorption process and first-order elimination
Abdelwahab, 2020 [79]	NONMEM	Body size, CL Body size, V	Pregnant women with TB/HIV (29), South Africa	One-compartment disposition, first-order elimination and transit compartment absorption
Ethambutol
Peloquin, 1999 [72]	ADAPTII		Adult healthy volunteers (14)	Two-compartment, lag time
Zhu, 2004 [73]	NPEM2	Serum creatinine, half life	Patients with TB, adults (56) and children (14), USA	One-compartment
Jönsson, 2011 [74]	NONMEM	HIV, F Body weight, CL Body weight, V	Patients with TB, adults (189), South Africa	Two-compartment, transit compartment absorption
Hall, 2012 [75]	ADAPT5	Weight, CL	Adults, healthy volunteers (18)	Two-compartment
Hiruy, 2015 [181]	ADAPT5	–	Patients with TB (children 31), South Africa	One-compartment
Denti, 2015 [76]	NONMEM	Age, CL Fat-free mass, CL Fat-free mass, V	Patients with TB, adults (98), Tanzania	Two-compartment, transit compartment absorption
Horita, 2018 [30]	NONMEM	HIV, V1/F Body weight, CL Body weight, V	Patients with TB, children (113), Ghana	Two-compartment, lag time
Mehta, 2019 [77]	Phoenix NLME	ART initiation, F	Adults with co-infected TB/HIV (40), Botswana	Two-compartment, lag time
Sundell, 2020 [78]	NONMEM	CYP1A2, F Body weight, CL Body weight, V	Patients with TB/HIV (63), adults, Rwanda	One-compartment, transit compartment absorption
Abdelwahab, 2020 [79]	NONMEM	Body size, CL Body size, V	Patients with TB/HIV (18), adults, pregnant women, South Africa	Two-compartment, transit compartment absorption
Levofloxacin
Peloquin, 2008 [198]	PASTRX, NPEM2, USC*PACK	–	Patients with TB (10), Brazil	One-compartment
van den Elsen, 2018 [89]	MWPharm	–	Patients with TB (30), Belarus	One-compartment with lag time
Denti, 2018 [199]	NONMEM	Body weight, CL Body weight, V	Patients with MDR-TB (children, 109), South Africa	Two-compartment, disposition kinetics, first-order elimination and absorption
Al-Shaer, 2019 [200]	Monolix	Sex, V Weight, V CrCL, CL	Patients with TB (108), Brazil, Georgia, Bangladesh, USA	One-compartment, first-order absorption and elimination
Moxifloxacin
Peloquin, 2008 [198]	PASTRX, NPEM2, USC*PACK	–	Patients with TB (9), Brazil	One-compartment
Pranger, 2011 [201]	MWPharm	–	Patients with TB (21), the Netherlands	One-compartment with first-order absorption and without lag time
Zvada, 2014 [202]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with TB (241), Zimbabwe, South Africa	Two-compartment, with first-order elimination and transit absorption compartments
Chang, 2017 [203]	NONMEM	Body weight, CL	Patients with MDR-TB (14), South-Korea	One-compartment, first-order absorption
van den Elsen, 2019 [18]	MWPharm	–	Patients with TB (77), the Netherlands	One-compartment with lag time
Al-Shaer, 2019 [200]	Monolix	–	Patients with TB (70), Brazil, Georgia, Bangladesh, USA	One-compartment, first-order absorption and elimination
Bedaquiline
McLeay, 2014 [99]	NONMEM	Study, F Black race, subject status, and DS-TB, CL/F Sex, Vc/F	Healthy adults and patients with TB (480)	Four-compartment disposition, absorption described by a dual zero-order input function
Svensson, 2016 [98]	NONMEM	Body weight, CL Albumin, CL Albumin, fraction of bedaquiline metabolized to M2	Patients with TB (335)	Three-compartment bedaquiline and one-compartment for metabolite M2
Linezolid
Alffenaar, 2010 [204]	MWPharm	–	Patients with MDR-TB (14), the Netherlands	One-compartment with first-order absorption pharmacokinetics without lag time
Kamp, 2017 [115]	MWPharm	–	Patients with MDR-TB (56), the Netherlands	One-compartment
Garcia-Prats, 2019 [205]	NONMEM	Body weight, CL Body weight, V	Patients with TB (children, 48), South Africa	One-compartment
Alghamdi, 2020 [206]
Clofazimine
Nix, 2004 [119]	NONMEM	–	Healthy adults (16), USA	One-compartment
Abdelwahab, 2020 [120]	NONMEM	Body weight, CL Fat-free mass, V	Patients with TB (139), South Africa	Three-compartment, first-order elimination and absorption in transit compartments
Faraj, 2020 [207]	NONMEM	–	Patients with TB (15), South Africa	Two-compartment disposition, first-order absorption and elimination
Cycloserine/terizidone
Chang, 2017 [203]	NONMEM	–	Patients with MDR-TB (14), South-Korea	One-compartment, first-order absorption
Mulubwa, 2019 [208]	Monolix	Albumin, V	Patients with TB (39), South Africa	One-compartment, modified transit compartment for terizodone absorption
Alghamdi, 2019 [124]	Monolix	Healthy/TB, CL CrCL, CL Body weight, V	Patients with TB (235), healthy adults (12), Georgia, Bangladesh, USA	One-compartment, with a first-order absorption and lag phase
van der Galiën, 2020 [125]	MWPharm	–	Patients with TB (17), Belarus	One-compartment, first-order absorption without lag time
Chirehwa, 2020 [209]	NONMEM	Fat-free mass, CL Fat-free mass, V	Patients with TB (132), South Africa	One-compartment disposition, with non-renal and renal clearance
Pretomanid
Lyons, 2018 [210]	GNU MCSim modeling and simulation suite	Body weight, CL Body weight, V	Patients with TB, adults (102), South Africa	One-compartment, first-order absorption and elimination and a sigmoidal bioavailability dependent on dose, time, and the pre-dose fed state
Salinger, 2019 [211]	NONMEM	Body weight, CL Body weight, V	Healthy adults and patients with TB (1054)	One-compartment
Amikacin
Dijkstra, 2015 [156]	MWPharm	Sex, V	Patients with TB (20), the Netherlands	One compartment
Ethionamide/prothionamide
Zhu, 2002 [162]	PASTRX, NPEM2, USC*PACK	–	Patients with TB (55), USA	One-compartment
Nyberg, 2020 [212]	NONMEM	Nasogastric administration HIV, F Rifampicin, CL Body weight, CL Body weight, V	Patients with TB (children, 110), South Africa	One-compartment disposition model with first-order elimination and a transit compartment absorption
Al-Shaer, 2020 [213]	Pmetrics	–	Patients with TB (94), Bangladesh, USA	One-compartment, first-order elimination and absorption lag time
Para-aminosalicylic acid
de Kock, 2014 [167]	NONMEM	Efavirenz, CL	Patients with TB (73), South Africa	One-compartment disposition, absorption described by a 3-transit compartment
Chang, 2017 [203]	NONMEM	–	Patients with MDR-TB (14), South-Korea	One-compartment with first-order absorption
Abulfathi 2020 [214]	NONMEM Modification of de Kock, 2014 [167]	Body weight, CL Body weight, V	Patients with TB (73), South Africa	One-compartment disposition with 3-transit absorption compartments

AIDS acquired immunodeficiency syndrome, CL clearance, CL/F apparent clearance, CrCL creatinine clearance, CYP cytochrome P450, DS drug-susceptible, F bioavailability, FDC fixed dose combination, HIV human immunodeficiency virus, IT2B iterative two-stage Bayesian procedure, Ke elimination rate constant, MDR multi-drug resistant, MTT mean transit time, NAT2 N-acetyl transferase 2, SDF single dose formulation, TB tuberculosis, V volume of distribution, V/F apparent central volume of distribution

Table 2

PK parameters and TDM of anti-tuberculosis drugs

Drug	DoseA	PK/PD target for efficacyB	AUC (mg h/L)B	Cmax (mg/L)C	MIC critical concentration (range) (mg/L)D [13]	TDM indi-catedE	LSSF (h)
INH	5 mg/kg	AUC/MIC > 567 (lungs) [22]	52 [31] <21.78 (T) [215]	8.8 [31] 3–6 [26]		E: Yes T: Yes	1, 2.5, 6 h [33] 1, 6, 8 h [34]
RIF	10 mg/kg	AUC/MIC > 271 [36] AUC/MIC: 435–683 [215]	38.7 [45] 13 [31]	5.79 [45] 6.6 [31] 8–24 [26]		E: Yes T: Yes	1, 3, 8 h [14] 2, 4 h [47]
PZA	25–35 mg/kg	AUC/MIC > 8.42 [215]	363 [31]	58.3 [31] 20–60 [26]		E: Yes T: Yes	0, 2, 6 h [33] 0, 5, 8 h [34]
EMB	25 mg/kg	AUC/MIC >119 [22] T>MIC (R) [67]		2–6 [26]		E: No T: Yes	0, 2.5, 6 h [33] 2, 4, 8 h [34]
LFX	750–1000 mg	AUC/MIC > 146 [83] AUC/MIC > 320 (R) [83]	110 (85–200)G [89, 198, 200]	10 (8–15)G [89, 198]	LJ: 2 (0.5-12) 7H10: 1 (0.06-8) 7H11: ND (0.06-32) MGIT: 1 (0.12-16)	E: Yes T: No	0, 5 h [89]
MFX	400 mg [53]	fAUC/MIC > 42 fAUC/MIC > 53 (R) [82]	35 (10–80)G [18, 198, 200, 202]	3.5 (2–6)G [18, 198, 202]	LJ: 1 (0.12-8) 7H10: 0.5 (0.02-8) 7H11: 0.5 (0.06-8) MGIT: 0.25 (0.06-8)	E: Yes T: No	0, 1.5, 6 h [33] 0, 6 h [18]
BDQ [93, 216]	400 mg QD for 14 days, 200 mg TW	AUC0–168 h/MIC or Cavg/MIC [98]	AUC0–168h: 187 (53–689)H [98]		7H10: ND (0.008-3.2) 7H11: 0.25 (0.008-0.5) MGIT: 1 (0.03-4)	E: YesJ T: Yes (M2)	0 h [98]
LZD [217, 218]	600 mg	fAUC/MIC: 119 [103, 105] Cmin < 2 mg/L (T) [108]	100 (107.5 ± 30.16) [103]	12–26 [26]	7H10: 1 (0.06-4) 7H11: 1 (0.06-32) MGIT: 1 (0.12-16)	E: Yes T: Yes	0, 2 h [115]
CFZ [121, 122]	100 mg			0.5–2.0 [26]	7H10: ND (0.06-1) 7H11: ND (0.12) MGIT: 1 (0.12-5)	E: No T: No
CS/TZ [203, 219]	250–750 mg	T > MIC 30% [126]		20–35 [26, 127, 128]	LJ: ND (7.5–60) 7H10: ND (3.75–32) 7H11: ND (7.5–60) MGIT: ND (4–64)	E: Yes T: Yes	4 h [125]
DLM	100 mg BID		7.9 [132]	0.41 [132]	7H10: ND (0.006–0.05) 7H11: 0.016 (0.001–0.12) MGIT: 0.06 (0.002–0.06)
ERT	1000 mg [220] 2000 mg [150]	fT > MIC 40% [150]					1, 5 h [16]
AM	15–20 mg/kg [53] 6.5 mg/kg [155]	Cmax/MIC > 75 [151,221] AUC/MIC > 103 [151, 152] Cmax/MIC > 20 [155]	568 [221] 113 (49–232)H [155]	67 [221] 46 (26–54)HK [153] 29.3 (11.0–72.5)H [155]	LJ: 30 (2–128) 7H10: 2 (0.25–160) 7H11: ND (0.25–64) MGIT: 1 (1–80)	E: Yes T: Yes	1, 4 h [156]
S	12–18 mg/kg [53]		197 ± 26L [222]	44 (33–55)HK [153] 42.0 ± 10.8LM [222]	LJ: 4 7H10: 2 7H11: 2 MGIT: 1	E: Yes T: Yes	1, 6 h [222]
ETA	250–500 mg	AUC/MIC > 56.2 [159] fAUC/MIC 42 [213]		1–5 [26, 162]	LJ: 40 7H10: 5 7H11: 10 MGIT: 5
PAS	4000 mg	fCmin > 1 mg/L [167]		20–60

AM amikacin, AUC area under the concentration–time curve from time 0–24 h, BDQ bedaquiline, BID twice daily CFZ clofazimine, C maximum plasma concentration, CS/TZ cyloserine/terizodone, DLM delamanid, EMB ethambutol, ERT ertapenem, ETA ethionamide, h hours, INH isoniazid, LFX levofloxacin, LZD linezolid, MFX moxifloxacin, MIC minimum inhibitory concentration, PAS para-aminosalicylic acid, PD pharmacodynamic, PK pharmacokinetic, PZA pyrazinamide, TDM therapeutic drug monitoring, R resistance, RIF rifampicin, S streptomycin, TW three times weekly

ADose is once daily, unless otherwise specified

BAUC0–24h in steady state, unless otherwise specified

CCmax in steady state, unless otherwise specified

DLJ—Löwenstein–Jensen medium, 7H10—Middlebrook 7H10 medium, 7H11—Middlebrook 7H11 medium, MGIT—BACTEC™ Mycobacterial Growth Indicator Tube™ 960, ND—not determined

ETDM indicated for E—efficacy or T—toxicity

FLSS—limited sampling strategy

GMean (normal range)

HMedian (range)

J(Selected cases) [100]

KCmax back calculated to end of intravenous infusion [153]

LMean ± standard deviation

MCmax 1 h after intramuscular injection of 1 g of streptomycin [222]

We used a non-systematic literature search in PubMed using a combination of the following keywords: tuberculosis, [drug name], pharmacokinetics, pharmacodynamics, sampling, and concentration. Case reports, abstracts, and posters were excluded. Included articles were limited to the English language. References of included articles were scanned for possible relevant articles. The search was last performed on 4 August, 2020.

Isoniazid

Preclinical Pharmacokinetics/Pharmacodynamics

Isoniazid is active against Mycobacterium tuberculosis (MTB), inhibiting the mycolic acid synthesis, which affects cell wall synthesis, resulting in a bactericidal effect [19]. Furthermore, isoniazid has an effect on DNA biosynthesis and has it has been shown that early bacterial activity (EBA) on the first 2 days of therapy is related to the acetylator type and 600 mg achieves the highest EBA [19, 20]. Hollow fiber infection models showed that isoniazid has bactericidal activity that is driven by AUC0–24 h/MIC [21, 22]. Studies in the mouse confirm this bactericidal activity, but isoniazid appears to be a weak sterilizing drug [23]. In the mouse, the activity of isoniazid appeared to be more concentration than time dependent, with Cmax/MIC and AUC0–24 h/MIC having a strong association with activity [24].

Clinical Pharmacokinetics/Pharmacodynamics

Isoniazid is fairly hydrophilic, with moderate clinical plasma concentrations 2 h after ingestion (C2) of 3–6 mg/L after a 300-mg dose, and 9–15 mg/L after a 900-mg dose [25, 26]. Isoniazid is unable to penetrate and accumulate in caseous granulomas [10]. In healthy volunteers and patients with TB starting on therapy, the administration of isoniazid with a high-fat meal modestly reduced Cmax and bioavailability [27, 28]. Both orange juice and aluminium–magnesium antacid produced little effect on the bioavailability of isoniazid [27]. Isoniazid metabolism is genetically determined based on N-acetyl transferase 2 activity [25, 27]. It has been proposed that patients be split between having none, one or two rapid N-acetyl transferase 2 alleles and that dosing could be then adapted to 2.5 (slow acetylators), 5.0, and 7.5 mg/kg (fast acetylators), respectively [29]. Pharmacokinetic data for isoniazid in adults and children with TB suggested that low and variable concentrations are common [30, 31]. In 142 drug-susceptible patients with TB, Pasipanodya et al. showed that to achieve a favorable outcome, in order of importance, AUC0–24 h of pyrazinamide, rifampicin, and isoniazid should be over 363, 13, and 52 mg h/L respectively. Low Cmax of rifampicin (< 6.6 mg/L) and isoniazid (< 8.8 mg/L) preceded all (three) cases of acquired drug resistance [31].

Population Pharmacokinetics, TDM, and Dosing

Standardized doses may not be adequate for isoniazid, especially given the differences between fast and slow acetylators. Therapeutic drug monitoring could be used to rapidly find the correct dose for each patient [32].

Twelve population PK models were found from the time period of 1997–2019 (Table 1). Most of the models were built with NONMEM and included the acetylation status/N-acetyl transferase 2 genotype as a covariate on clearance.

Limited sampling strategies were developed for all four first-line TB drugs and moxifloxacin in an ethnically heterogenous population, using intensive PK sampling after the intake of drugs on an empty stomach and after more than 2 weeks of therapy [33]. For isoniazid, LSSs of 1, 2.5, and 6 h post-dose performed best [33]. Recently, several LSSs were proposed for isoniazid alone or in combination with other first-line drugs. Limited sampling strategies were derived from a study with intense sampling of 20 patients conducted in Indonesia, who started on TB medication [34]. Both studies used the best subset selection, multiple linear regression to calculate LSS and the chosen LSS performed best at estimating the AUC0–24 h.

Rifampicin

Rifampicin binds to and inhibits the action of the DNA-dependent RNA polymerase of mycobacteria. It has both bactericidal and sterilizing activity. In the past decades, it was assumed that the efficacy of rifampicin is associated with Cmax/MIC, but recent preclinical studies have established that the microbial killing can also be explained by AUC/MIC [35, 36]. Given the short half-life of rifampin of around 2 h at steady state, there is a high correlation between Cmax and AUC. Moreover, in one of these studies, it was well presented that in a mouse model with dose fractionation and ranging, the PK/PD parameter that correlated best with a decrease of bacterial counts was AUC/MIC [36].

Rifampicin is preferably taken on an empty stomach, as intake with food reduces Cmax and causes a modest decrease in AUC0–24 h [28, 37]. However, in the case of gastrointestinal adverse effects, concomitant intake with a light meal is recommended to prevent or alleviate these effects. The distribution of rifampicin is fast and includes all fluids and organs, but only about 5% of rifampicin reaches the cerebrospinal fluid [38].

Repeated daily administration of rifampicin results in a decrease in exposure and half-life of the drug, owing to an increased rate of drug clearance caused by auto-induction [39]. This is separate from induction of the cytochrome P450 system by rifampicin (see below). Increasing the dose of rifampicin results in a more than proportional increase of exposure in plasma (non-linear pharmacokinetics). Rifampicin is excreted mainly through the bile. Urinary excretion is about 10–15% of biliary excretion but increases with the dose.

Rifampicin is a potent inducer of several phase I and II metabolic enzymes and drug transporter proteins, often resulting in a decrease in exposure of concomitantly used drugs [40]. This is especially true for cytochrome P450 3A4, a major drug-metabolizing pathway. Significant induction after starting rifampicin is reached after a few doses and is complete after a week [40]. Baseline enzyme activity after discontinuing rifampicin is attained in about 2 weeks.

The drug is currently used in a dose of 10 mg/kg daily, but this is at the lower end of the dose–response curve. A dose of 35 mg/kg daily, resulting in a ten-fold higher AUC0–24 h value in plasma, is safe and tolerable for 3 months [41, 42]. Moreover, higher dosing has shown a significant relationship between exposure to rifampicin and EBA [43]. Higher rifampicin exposures also reduce the time to culture conversion in pulmonary TB [44]. A meta-analysis on the exposure of rifampicin reported that at least 25-mg/kg dosing is required to achieve PK/PD targets [45].

Nineteen population PK models for rifampicin in adult humans with fully re-implementable parameter values could be identified in the time period from 1997 up to 2018 (Table 1). These models differed with respect to the number of compartments, the inclusion of components for non-linearity and for auto-induction, which are both typical to rifampicin. Only one model was based on higher doses of rifampicin up to 40 mg/kg [46]. Most of the models were built in NONMEM.

Limited sampling strategies were developed using the best subset selection, multiple linear regression for all four first-line TB drugs and moxifloxacin in an ethnically heterogenous population, using intensive PK sampling after the intake of drugs on an empty stomach and after more than 2 weeks of therapy [33]. Limited sampling at various fixed sampling points up to 6 h enabled an accurate and precise prediction of AUC0–24 h of all drugs separately and simultaneously. The limited sampling formula called for 2-, 4-, and 6-h samples [33]. Another study also applied a linear regression approach, but focused on the prediction of the AUC0–24 h of rifampicin and other TB drugs during the first 3 days of treatment [34]. In the homogenous Indonesian population studied, limited sampling at 2, 4, and 8 h post-dose performed best to predict AUC0–24 h values of all first-line TB drugs.

Sturkenboom et al. were the first to use a model-based (maximum a posteriori Bayesian fitting) approach to derive a LSS for rifampicin [14]. They extended the population used by Magis-Escurra et al. [33], which took rifampicin on an empty stomach, to patients who ingested the drug with a light meal and derived a population PK model in MWPharm (Mediware, Zuidhorn, the Netherlands). This study showed that rifampin AUC0–24 h could be predicted with acceptable bias and precision with optimal sampling at time points 1, 3, and 8 h post-dose. Furthermore, a sampling strategy using blood collection at convenient 2- and 4-h sampling times has been shown in another study to be the most suitable [47]. This NONMEM model was successfully implemented in the InsightRX (San Francisco, CA, USA) precision dosing platform.

Pyrazinamide

Pyrazinamide is a prodrug, which is activated to pyrazinoic acid inside the MTB by the bacterial pyrazinamidase [48]. It is mainly active against non-replicating bacilli, especially at low pH [49], which allows for accumulation of pyrazinoic acid inside the MTB [50]. In pre-clinical hollow fiber infection models, the sterilizing effect of pyrazinamide was most closely related to the AUC0–24 h/MIC ratio [51]. The MIC determination is challenging, owing to the difficulty of reproducibly growing MTB at low pH across laboratories.

Pyrazinamide is used both against drug-sensitive [52] and resistant TB [53] and it is often dosed 20–30 mg/kg bodyweight. It should be noted that the original British Medical Research Council studies used fixed dosing of 1500 mg for patients less than 46 kg, and 2000 mg for larger patients, with an average dose of 35 mg/kg. A large number of PK studies have been published on pyrazinamide at the 20–30 mg/kg dose: its pharmacokinetics is described by a one-compartment open model [54], it reaches Cmax approximately 2 h post-dose and with a terminal half-life of approximately 9 h it only moderately accumulates, reaching steady state in 2–3 days. It does not significantly bind to plasma proteins [55] and it is metabolized by liver deaminase [56]. A recent review summarizing exposures in patients (human immunodeficiency virus positive/negative) reported median AUC0–24 h in the range of 250–450 mg h/L and Cmax in the range of 25–55 mg/L [57]. Several reports indicated that the current weight-banded dosing approach achieves lower concentrations in patients with lower weight [58-61].

In clinical studies in patients with drug-susceptible TB, worse outcomes were observed in patients with Cmax < 35 mg/L, while a positive association between AUC0–24h > 363 mg h/L and long-term TB treatment outcome has been reported [31, 62]. The main adverse reactions to pyrazinamide are arthralgia and hepatotoxicity [63]. Doses higher than the current doses have previously been used, and there is a debate regarding increasing toxicity with increasing doses [64].

Nine population PK models were found, from which three were made with NONMEM, three with Monolix, one with ADAPT5, and one with NPEM2 (Table 1). Only one model included a covariate sex on clearance [65]. Although a robust target for efficacy has not been established and safety is not of concern at the current levels, a TDM strategy could be useful to identify patients with low exposure using, for example, LSSs of 0, 2, and 6 h or 2, 4, and 8 h post-dose [33, 34].

Ethambutol

The precise mode of action of ethambutol is not fully understood, but it appears to inhibit mycobacterial cell wall arabino-transferases, leading to depletion of arabinogalactan and lipoarabinomannan. In an early study by Dickinson et al. in guinea pigs, the effects of ethambutol were concentration dependent (Cmax/MIC) [66]. More recent experiments in the hollow fiber infection model showed that the microbial kill to ethambutol was linked to AUC0–24 h/MIC, although in some experiments Cmax/MIC could also have explained the kill, whereas T > MIC was relevant for suppressing resistance [67].

It often is used in a dose of 15–20 mg/kg daily, with a maximum of 25 mg/kg daily. Higher doses and exposures, especially in patients with decreased renal function, are associated with a higher incidence of retrobulbar (optic) neuritis [68]. The EBA of the 15–20 mg/kg dose is negligible and the drug has very little sterilizing activity. However, ethambutol is added to other first-line TB drugs during the first weeks of treatment to prevent rifampin resistance in cases where there may be unrecognized isoniazid resistance. A clinical trial in Indian patients with TB showed that microbiological response was AUC/MIC driven [69, 70].

Ethambutol can be taken with or without food, as food modestly reduces Cmax but does not affect the AUC0–24 h of the drug [28, 37]. It has rapid good distribution throughout the body (including in large lung lesions), but it only penetrates the cerebrospinal fluid in the presence of inflammation, probably having a minor contribution to the treatment of TB meningitis [71]. A recent evaluation using ultrafiltration showed an average protein binding of 12% [55]. Ethambutol is largely excreted unchanged by the kidneys. Patients with a creatinine clearance less than 30 mL/min are at risk for accumulation of the drug and resulting retrobulbar neuritis. They should receive a longer interval between doses with three times a week administration of the same 15–20 mg/kg dose. In persons with good renal function, ethambutol shows a biphasic decline in plasma concentration with a terminal half-life of about 10–12 h. The drug shows no relevant drug interactions, apart from a reduction in its Cmax caused by aluminum-containing antacids [72].

Traditionally, limited sampling of ethambutol aimed at ‘catching’ and estimating the Cmax of the drug in an individual by sampling at 2 and 6 h after dosing. Recently developed LSSs aim to predict the AUC0–24 h for ethambutol as well. As described earlier, Magis-Escurra et al. developed an LSS using best subset selection, multiple linear regression for all first-line TB drugs (including ethambutol) separately and simultaneously [33]. Saktiawati et al. also applied a linear regression approach, but focused on prediction of the AUC0–24 h of first-line TB drugs during the first 3 days of treatment, based on data from Indonesian patients with TB [34]. Horita et al. were the only investigators to use a Bayesian approach to estimate AUC0–24 h of ethambutol in children [30].

Ten population PK models for ethambutol could be identified (Table 1) [30, 72–79]. These models differ with respect to the number of compartments, the inclusion of a lag time, or transit compartment absorption. Most of the models were built in NONMEM. It is important to note that a decreased renal function may be the most relevant indication for TDM of ethambutol, but relatively few patients with renal dysfunction were included in the various cohorts. This means that the effect of this covariate on clearance has not been adequately evaluated in population PK models. Similarly, only one model was derived in obese patients taking ethambutol [75].

Levofloxacin/Moxifloxacin

Fluoroquinolones are amongst the most potent drugs against MTB. They produce their rapid killing effect by inhibiting the bacterial DNA replication and transcription through affecting the type II topoisomerases [80].

The efficacy of fluoroquinolones is correlated with the ratio of the fAUC0–24 h/MIC [81-83]. In addition, an 800-mg daily dose for moxifloxacin is suggested to achieve maximal kill and suppress resistance [84, 85].

Therapeutic drug monitoring of moxifloxacin and levofloxacin is proposed because of high PK variability and a substantial proportion of patients showing low drug exposure [86-88]. Fluoroquinolones are well tolerated and adverse effects do not require TDM (e.g., QT interval prolongation by moxifloxacin). Therapeutic drug monitoring is helpful to ensure adequate exposure after dose adjustments of levofloxacin due to kidney failure or in the case of drug–drug interactions. For instance, moxifloxacin exposure is approximately 25% decreased if concomitantly used with rifampicin owing to an increase of moxifloxacin clearance [18]. Four population PK models were identified for levofloxacin and six for moxifloxacin (Table 1). The models were made with NONMEM, MWPharm, NPEM2, and Monolix and the majority were one-compartment models. To reduce the burden and costs of TDM, LSSs can be applied. The under the concentration–time curve from time 0 to 24 h can be adequately estimated using a population PK model and two optimally timed samples for levofloxacin (0 and 5 h) and for moxifloxacin (0 and 6 h, either with or without rifampicin) [18, 89].

Bedaquiline

Bedaquiline is a diarylquinoline that targets mycobacterial ATP synthase, disrupting the bacteria’s energy metabolism [90]. Dose-fractionation experiments in mice identified AUC/MIC as the main driver of bactericidal effect [91].

The standard dosing regimen in the treatment of MDR-TB consists of a 2-week loading phase with 400 mg daily followed by 22 weeks with 200 mg dosed three times per week. An alternative dosing regimen with 200 mg daily during the first 8 weeks, thereafter 100 mg daily, has been utilized in recent clinical trials [92]. Uptake of bedaquiline is increased when given together with food and maximal concentrations are expected 4–6 h after dosing. The binding to plasma proteins is very high, more than 99.9%, and the distribution to tissues is extensive, resulting in an extremely long terminal half-life [93]. One case report showed undetectably low bedaquiline concentrations in cerebrospinal fluid [94].

Bedaquiline is mainly hepatically cleared. Cytochrome P450 3A4 (primarily) transforms bedaquiline into the N-monodesmethyl metabolite M2, which in turn is metabolized in the same way to M3 [93, 95].

A model-based PK/PD analysis of phase II data found a weekly average concentration (analogous to weekly AUC) to be the best driver and estimated that a concentration of 1.42 mg/L would give half of the maximal possible effect [96]. The relationship has also been confirmed in a separate study [97]. With the standard regimen typical (mean ± one standard deviation) maximal, average, and trough concentrations at day 14 are 1.6–3.2 mg/L, 0.96–2.1 mg/L, and 0.44–1.4 mg/L, respectively [98]. At the end of the continuation phase, the corresponding values are 0.9–2.1 mg/L, 0.41–1.2 mg/L, and 0.26–0.91 mg/L [98].

Pharmacokinetics, TDM, and Dosing

There are two published population PK models of bedaquiline in patients with drug-resistant pulmonary TB, partly based on the same data and both developed in NONMEM (Table 1). The first only describes bedaquiline [99], while the second also incorporates the main metabolite M2 [98]. There are no published evaluations of TDM sampling strategies for bedaquiline. The time on treatment needs to be considered, accounting for the loading phase and the accumulation. Assessing the Cmin at the end of the loading phase is a practically feasible option. A retrospective study described potential risk factors of low and high bedaquiline concentrations, where administration of food, drug–drug interactions, gastrointestinal complaints, lower body weight, and age > 70 years were among the named factors [100]. In addition, black race has been connected to approximately 50% clearance [99]. Therapeutic drug monitoring could be of benefit for specific patients using bedaquiline; however, standardized drug susceptibility testing must first be developed [101]. The M2 metabolite is the main driver of side effects such as QT prolongation and phospholipidosis [102].

Linezolid

Linezolid exerts its antibacterial effect by binding 70S ribosomal components and disrupting the initiation of protein synthesis. A hollow fiber infection model study has indicated optimal mycobacterial kill to be associated with fAUC0–24 h/MIC [103]. In murine models, its sterilizing activity is dose related and can occur within 2 months of effective combination therapy, leading some investigators to consider short high-dose therapy followed by discontinuation or intermittent dosing to retain efficacy whilst reducing toxicity [104, 105].

The bioavailability after oral administration of linezolid is nearly 100%, and it shows good tissue penetration into tuberculous lung cavities caseating granulomas and cerebrospinal fluid [10, 106]. The standard adult dose (600 mg twice daily) is associated with myelosuppression, lactic acidosis, and neurotoxicity during long-term use. Consequently, a dose of 600 mg, or occasionally 300 mg, once daily is often recommended for patients with TB [104, 107].

The PK/PD target for myelotoxicity is suggested as a Cmin of 2 mg/L [108]. While meta-analyses and simulations from some of these data indicate high attainment of putative efficacy targets at 600 mg once or twice daily, Cmin regularly exceeds the toxicity threshold with twice-daily dosing [109, 110]. One study reported that higher plasma linezolid exposure and greater toxicity risk are associated with advancing age and lower weight [111]. It has been reported that 300 mg twice daily might be effective [104]. Studies looking into 300-mg daily dosing have shown promising results in some patients [112, 113]; however, until larger trials show results, 600 mg has been suggested to keep the balance between efficacy and toxicity [105]. No association between human immunodeficiency virus or concurrent anti-retroviral therapy and linezolid exposure has been described [111]. An approach of short-duration higher dose linezolid to maximize efficacy whilst avoiding toxicities associated with cumulative dose requires further evaluation; however, an interim analysis of the Nix-TB trial showed a significant amount of patients experiencing linezolid toxicity [114].

The narrow therapeutic window has prompted suggestion that TDM might be beneficial. An LSS consisting of 0 and 2 h has been described to predict the plasma AUC0–24 h with acceptable bias and precision [115]. Three population PK linezolid models were identified (Table 1), which were all one-compartment models and two were developed with MWPharm and one with NONMEM.

Clofazimine

Studies in the mouse model show that clofazimine did not have EBA, but subsequently demonstrated potent, dose- and concentration-independent bactericidal activity [116]. The authors suggested that much lower doses could be used effectively for TB and that clofazimine exhibits slow elimination [116]. In another mouse model, the activity of clofazimine was confirmed to be best when administered with other first-line anti-TB drugs [117].

Most clinical studies have used 100-mg daily doses, thus the literature lacks human dose-ranging data for TB. Very limited PK/PD data are available for clofazimine, with most only available as meeting abstracts.

Clofazimine is highly lipophilic, leading to high accumulation in fat tissues and relatively low serum concentrations (0.7–1.0 mg/L) [118]. Clofazimine is unable to penetrate and accumulate in caseous granulomas, but it does accumulate in the highly cellular peripheral zone of the granuloma, consisting of macrophages, epithelioid macrophages, and lymphocytes [118]. In healthy volunteers, the administration of clofazimine with a high-fat meal provided the greatest bioavailability. However, bioavailability was associated with high inter- and intra-subject variability. Both orange juice and an aluminum-magnesium antacid reduced a mean bioavailability of clofazimine [119].

Overall, three NONMEM population PK models were identified (Table 1). A recent population PK model on NONMEM described clofazimine disposition being strongly affected by body fat, thus clofazimine showed lower plasma concentrations in women [120]. Additionally, it was suggested that patients with extreme body composition might need to have individualized dosing and a high-fat content might require longer loading periods [120]. As mentioned, specific PK/PD targets for clofazimine are scarce. Clofazimine is suggested to be sampled 2–3 h after the dose and 6 h to assess delayed absorption, whereas Cmax is expected at 0.5–2.0 mg/L [26, 121, 122].

Cycloserine/Terizidone

Cycloserine is an amino-acid derivative, which suppresses the growth of MTB by inhibiting the enzymes that produce peptidoglycan causing changes in the cell wall [123]. Terizodone contains two cycloserine molecules. Hollow fiber infection models have proposed that T > MIC predicts efficacy best [124-126].

Cycloserine has been suggested to be used in patients with rifampicin-resistant TB as a Group B drug. It may also be of value in the therapy of TB meningitis owing to its penetration into cerebrospinal fluid [126].

As cycloserine has a long half-life, different Cmax ranges have been reported. To determine exposure, the first measurement should be done after 3–4 days of therapy and Cmax is expected to be around 20–35 mg/L [26, 127, 128].

Two population PK models have used the preclinical established PK/PK index T > MIC over 30% model [125, 129]. Five cycloserine/terizidone population PK models were identified and all of these were one-compartment models made with different software, Monolix, NONMEM, and MWPharm (Table 1). Recently, a sampling strategy using blood collection at a convenient 4 h has been proposed in combination with a population PK model to estimate AUC0–24h obtaining a low relative error [125].

Delamanid

Preclinical

Delamanid is a nitroimidazole antibiotic with mechanisms of action including inhibition of mycolic acid synthesis and generation of nitrous oxide in anaerobic conditions [130]. In animal models, the activity of delamanid activity appears to be time dependent [130].

The licensed dose is 100 mg twice daily, for 24 weeks, but an alternative dosing regimen of 100 mg twice daily for 8 weeks followed by 200 mg once daily for 16 weeks was used in the recent phase III trial [131]. The main route of metabolism of delamanid is purported to be via albumin, with cytochrome P450 3A playing a minor role [132]. It has several metabolites, but DM-6705 circulates at the highest concentrations. Delamanid and DM-6705 have half-lives of 30–38 h and 121–425 h, respectively, and the latter is linked to the key, albeit modest, adverse effect of delamanid—prolongation of the cardiac QT interval [133]. The bioavailability of delamanid is enhanced two- to four-fold by food, exposures are less than dose proportional, the drug is highly protein bound (> 99.5%), and the Tmax is at 4 h [134, 135]. This drug appears to reach high concentrations in the brain; its utility in central nervous system TB is being explored [136]. Delamanid is not an inducer or inhibitor of cytochrome P450 enzymes, and drug–drug interaction liability is low. In human EBA studies, doses of 200–300 mg total daily dose produced higher microbiologic activity than 100 mg per day [134]; in a phase II trial, there was a trend towards better activity of delamanid 200 mg twice daily compared with 100 mg twice daily on solid medium [137].

Median delamanid AUC0–24 h after 56 days of dosing is 7.9 mg h/L (coefficient of variation 37.5%). Maximum plasma concentration values are 0.4 mg/L (coefficient of variation 40.5%) at 56 days [132, 136]. There are no clear target concentrations owing to a lack of clinical PK/PD data. It has been reported that delamanid acquires resistance quickly, thus optimization of therapy using TDM may help in reducing the acquired resistance [138].

Pretomanid

Pretomanid was introduced with the BPal regimen (bedaquiline and pretomanid with linezolid). An in vivo model presented activity comparable to isoniazid, where the suggested reasoning was lower bioavailability in necrotic lung granulomas [139]. In another mouse model, the addition of pretomanid showed bactericidal activity increase while preventing emergence of resistance to bedaquiline and duration of therapy [140]. Later, a synergistic three-drug combination, including pretomanid, was suggested in a time kill assay combined with population modeling [141].

Pretomanid has shown to have higher bioavailability when taken with food [142]. In healthy adults it has been shown that AUC and Cmax could be almost doubled [142]. Moreover, drug interactions have been described with rifampicin, lopinavir/ritonavir, and efavirenz—all of these reduce the AUC of pretomanid [143].

Two population PK models were identified, one in NONMEM and the other developed on GNU MCSim modeling and simulation suite (Table 1).

Meropenem/Imipenem-Cilastatin/Ertapenem

Meropenem, a carbapenem, is a member of the β-lactam class of antibiotics. It inhibits cell wall synthesis through inhibition of classical D, D-transpeptidases, and non-classical L, D transpeptidases [144]. Clavulanate (not available except co-formulated with amoxicillin) irreversibly inhibits MTB β-lactamase, potentiating the activity of meropenem. Ertapenem has shown to a have sterilizing effect against M. tuberculosis in a hollow fiber infection model; however, falsely high MIC values might have been reported because of ertapenem degradation in acidic conditions, which could have led to underuse of ertapenem in TB therapy [145, 146].

In an EBA study, meropenem given at a dose of 2 g three times daily intravenously (supplemented with twice-daily amoxicillin/clavulanate) produced measurable activity in patients with pulmonary TB [147]. Recently presented EBA dose-finding trial results suggest that there is a significant drop in microbiologic activity when the meropenem dose is reduced from 6 g per day (given as 2 g three times daily) to 3 g per day (with 1 g three times daily performing modestly better than 3 g once daily). Meropenem, together with amoxicillin-clavulanate, is used off-label for TB. Doses for other bacterial infections range from 1.5 to 6 g per day, given in divided doses to maximize the activity of this time-dependent drug. Most (around 70%) of an intravenous meropenem dose is excreted unchanged in the urine [148]. Protein binding is very low (2%). Ertapenem in a dose of 1 g once daily intramuscularly in combination with amoxicillin-clavulanic acid showed no EBA [149]. The lack of activity of this dose is in line with an earlier published study using the hollow fiber infection model indicating that a dose of at least 2 g once daily was required [146].

Mean meropenem AUC0–24 h given at 2 g three times daily is 573 mg h/L, with a Cmax of 133 mg/L; at that dose, with an MIC of 1 mg/L, T > MIC is appropriately 75% [134]. In the absence of a PK target that maximizes efficacy against slow-growing MTB, it seems reasonable to strive for these average PK values. Collecting samples for the PK pre-dose and mid-dosing interval is most informative. While carbapenems are generally well tolerated, there is an increased risk of neurotoxicity with Cmin > 16 mg/L (and the dose-limiting toxicity is chiefly diarrhea from the amoxicillin–clavulanate). Therapeutic drug monitoring of meropenem in TB could thus serve to help with individualized dosing of meropenem, achieving exposures similar to those seen with maximal EBA activity and avoiding toxic Cmin.

Ertapenem in a 2000-mg once-daily pharmacokinetic exploratory study showed that f40% of T > MIC was reached for most patients [150]. A PK model based on data from patients with MDR-TB proposed a limited sampling strategy of 1 h and 5 h [16].

Amikacin/Streptomycin

In a hollow fiber infection model for TB, the local amikacin Cmax/MIC ratio of 10 was determined to be the effective PK/PD index, closely followed by AUC/MIC [151, 152].

Aminoglycoside use is limited by both the parenteral administration and significant toxicity that is observed after prolonged use [153] and should only be used if drug susceptibility testing results confirm susceptibility and audiometry monitoring can be ensured [53]. A 10% probability of ototoxicity occurred with a threshold cumulative AUC > 87 g h/L [154]. It has been revealed that the probability of ototoxicity increased sharply starting after 6 months of amikacin therapy to near maximum at 9 months. Considering the poor penetration of aminoglycosides in lung tissue, this translates into a serum Cmax/MIC ratio of 70–90 or an AUC0–24 h/MIC ratio of 103 [151].

Therapeutic drug monitoring of aminoglycosides is indicated to ensure efficacy and should be combined with determination of MIC as the risk of ototoxicity can be reduced if the MIC is low enough to lower the (cumulative) dose and exposure [155]. A one-compartment population PK model has been reported for amikacin. This included a LSS of 1 h and 4 h suggested for predicting AUC0–24 h [156].

Ethionamide/Prothionamide

The thiomides ethionamide and its propyl-analog prothionamide are drugs with bactericidal activity against MTB [157]. They are considered interchangeable within anti-tubercular regimens [158]. In hollow fiber systems, the target exposure was identified as AUC0–24 h/MIC [159].

It is currently used for MDR-TB at 15–20 mg/kg daily (maximum 1000 mg) [53] and for drug-susceptible TB in children or TB meningitis. The plasma protein binding of ethionamide is approximately 30% [160] and it is metabolized by hepatic monooxygenases [161].

Three population PK ethionamide models were identified, from which all were one-compartment models using NONMEM and Pmetrics software (Table 1). Clinically, after administration of 250–500 mg of ethionamide, a Cmax in the range of 1–5 mg/L has been suggested [26]. A TDM strategy could identify patients with low concentrations by evaluating two samples at 2 h and 6 h with a Cmax of 1–5 mg/L [26, 162, 163].

P-Aminosalicylic Acid

The mechanism of action of P-amino salicylic acid (PAS) remains unknown but it is assumed that it interferes with bacterial folate synthesis [164].

Although one study described a bactericidal effect of PAS at high (20 g) once-daily dosing [165], it is generally regarded as a bacteriostatic agent. It is associated with considerable gastrointestinal intolerance and was quickly replaced by isoniazid, rifampicin, and pyrazinamide in first-line anti-TB therapy. It has recently been downgraded in the World Health Organization MDR-TB guidelines but retains a role as component of treatment for patients with extensively drug resistant TB (XDR-TB) and limited other options [53]. Current PAS dosing for adults is 150 mg/kg/day in two to four divided oral doses (usually 8–12 g/day), and 200–300 mg/kg/day in two to four divided doses for children. Absorption is improved by consumption with acidic food or yoghurt [166]. It attains its maximum serum concentration within about 2 h, and has a fairly short half-life [165]. Protein binding is between 50 and 60% [167]. Current granular formulations are designed to provide slow release, assuming a fCmin > 1 mg/L will be maintained [167]. Some authors have reported that a high Cmax may be more important for prevention of resistance to partner drugs [168].

Three NONMEM population PK models were identified, which were all one-compartment models (Table 1). A role for TDM in PAS dosing has not yet been explored.

Concluding Remarks

In this review, over 80 different population PK models for 14 anti-TB drugs are presented.

As different models are difficult to compare and because the true model cannot be identified, there is no such thing as the ‘best’ or most recommended model. This was already stated in 1976, in one of the most famous adages in pharmacokinetics: ‘all models are wrong, some are useful’ [169]. Our personal preference would be using a validated model and performing TDM with Bayesian estimation to most accurately estimate AUC for the individual patient and situation. This can be done using different modeling and Bayesian simulation software. A good overview of different available methods has recently been presented [170].

By including TDM as an opportunity for optimizing TB treatment in TB guidelines [53, 171, 172], the first step towards implementation of precision medicine has been made. The guidelines include indications and clinical conditions for which TDM may be useful. To be of practical use, guidance is required on how drug exposure should be evaluated and how dosages could be adjusted [173]. We therefore performed this review evaluating the available literature on pharmacokinetics and PK modeling of anti-TB drugs. For the first-line anti-TB drugs, it is clear that studies focussed mainly on rifampicin and isoniazid [25, 174–176]. Pharmacokinetic variability is substantial and can be explained by pharmacogenetic differences and other factors such as co-morbidities [45, 177, 178]. With respect to dose optimization, there is a clear rationale based on in vitro PK/PD and clinical studies. Unfortunately, a well-designed study comparing TDM with standard of care is lacking. Although higher dosages are currently being trailed for rifampicin, TDM will likely still have a role as PK variability is still high, which leads to the conclusion that a high dose does not automatically result in a high exposure in every patient.

For anti-TB drugs used for the treatment of M/XDR-TB, there is a significant knowledge gap for many of the drugs. Only for the fluoroquinolones, linezolid, and aminoglycosides does a substantial body of evidence support TDM. The new drugs such as bedaquiline, delamanid, and pretomanid lack this abundance of data but publications on acquired resistance, drug–drug interactions, and significant variability of bioavailability based on concomitant food intake already provide a rationale that TDM may be of use for the newer drugs as well. However, it needs to be mentioned again that a well-designed prospective study comparing TDM in M/XDR-TB is still lacking. As researchers, we ourselves often struggle with conducting prospective studies to show the value of TDM. Recently, a review by Märtson et al. identified knowledge gaps and provided guidance with an appropriate checklist and data elements that are suggested to take into account when designing and conducting high-quality TDM studies in the field of infectious diseases [179]. Briefly, the focus was on estimating drug exposure and assessing the susceptibility of the pathogen, but also suggestions on how to select study endpoints and clinical trial design were provided [179].

Logistical issues regarding TDM samples seem to be a hurdle that will be overcome in the next couple of years owing to the introduction of dried blood spot monitoring, point of care saliva tests, and urine testing [173, 180]. To support a physician in making informed dosing decisions in individual patients, easy-to-use model-informed precision dosing software has to be made freely accessible via mobile platforms such as smartphones and tablets. On-site patient covariates and drug concentrations can be used to simulate dosing regimens to maximize the likelihood to attain the therapeutic target. Such software is currently under development for drugs such as vancomycin but there is a clear need to expand the panel with anti-TB drugs.

As TB treatment duration is long, there is an urgent need to find biomarkers that predict clinical outcome. A combination of dose optimization based on measured drug concentration and pathogen susceptibility with a sensitive and responsive biomarker would be helpful to optimize treatment.

Since the introduction of TDM for TB over 3 decades ago, further development of TDM in TB next steps will again depend on academic and clinical initiatives supported by funding institutions such as EDCTP or the Bill & Melinda Gates Foundation, as the pharmaceutical industry is not interested in personalized prescribing for the novel compounds and the generic pharmaceutical industry has no role in drug development or treatment optimization. We recommend that researchers work closely together with guideline issuing bodies like the World Health Organization to ensure the evidence generated will be reviewed during guideline updates. With such an approach, TDM will become available within the next 5 years and can help to contribute to END TB.

Information on pharmacokinetics and pharmacodynamics of anti-tuberculosis drugs can help to optimize treatment.

Population pharmacokinetics and Bayesian dose adjustment can help to optimize dose adjustment.

There is still a significant knowledge gap for many of the tuberculosis drugs.