Tuberculosis drugs' distribution and emergence of resistance in patient's lung lesions: A mechanistic model and tool for regimen and dose optimization

Abstract

Background: The sites of mycobacterial infection in the lungs of tuberculosis (TB) patients have complex structures and poor vascularization, which obstructs drug distribution to these hard-to-reach and hard-to-treat disease sites, further leading to suboptimal drug concentrations, resulting in compromised TB treatment response and resistance development. Quantifying lesion-specific drug uptake and pharmacokinetics (PKs) in TB patients is necessary to optimize treatment regimens at all infection sites, to identify patients at risk, to improve existing regimens, and to advance development of novel regimens. Using drug-level data in plasma and from 9 distinct pulmonary lesion types (vascular, avascular, and mixed) obtained from 15 hard-to-treat TB patients who failed TB treatments and therefore underwent lung resection surgery, we quantified the distribution and the penetration of 7 major TB drugs at these sites, and we provide novel tools for treatment optimization.

Methods and findings: A total of 329 plasma- and 1,362 tissue-specific drug concentrations from 9 distinct lung lesion types were obtained according to optimal PK sampling schema from 15 patients (10 men, 5 women, aged 23 to 58) undergoing lung resection surgery (clinical study NCT00816426 performed in South Korea between 9 June 2010 and 24 June 2014). Seven major TB drugs (rifampin [RIF], isoniazid [INH], linezolid [LZD], moxifloxacin [MFX], clofazimine [CFZ], pyrazinamide [PZA], and kanamycin [KAN]) were quantified. We developed and evaluated a site-of-action mechanistic PK model using nonlinear mixed effects methodology. We quantified population- and patient-specific lesion/plasma ratios (RPLs), dynamics, and variability of drug uptake into each lesion for each drug. CFZ and MFX had higher drug exposures in lesions compared to plasma (median RPL 2.37, range across lesions 1.26-22.03); RIF, PZA, and LZD showed moderate yet suboptimal lesion penetration (median RPL 0.61, range 0.21-2.4), while INH and KAN showed poor tissue penetration (median RPL 0.4, range 0.03-0.73). Stochastic PK/pharmacodynamic (PD) simulations were carried out to evaluate current regimen combinations and dosing guidelines in distinct patient strata. Patients receiving standard doses of RIF and INH, who are of the lower range of exposure distribution, spent substantial periods (>12 h/d) below effective concentrations in hard-to-treat lesions, such as caseous lesions and cavities. Standard doses of INH (300 mg) and KAN (1,000 mg) did not reach therapeutic thresholds in most lesions for a majority of the population. Drugs and doses that did reach target exposure in most subjects include 400 mg MFX and 100 mg CFZ. Patients with cavitary lesions, irrespective of drug choice, have an increased likelihood of subtherapeutic concentrations, leading to a higher risk of resistance acquisition while on treatment. A limitation of this study was the small sample size of 15 patients, performed in a unique study population of TB patients who failed treatment and underwent lung resection surgery. These results still need further exploration and validation in larger and more diverse cohorts.

Conclusions: Our results suggest that the ability to reach and maintain therapeutic concentrations is both lesion and drug specific, indicating that stratifying patients based on disease extent, lesion types, and individual drug-susceptibility profiles may eventually be useful for guiding the selection of patient-tailored drug regimens and may lead to improved TB treatment outcomes. We provide a web-based tool to further explore this model and results at http://saviclab.org/tb-lesion/.

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1. Raw clinical data for each drug in each lesion type.

Concentration–time profiles are shown for 9 lesions and 7 drugs by respective panel. Plasma concentrations over time for each individual were measured at multiple time points after the time of drug administration and before lung resection and are shown as individual lines of different colors. Lesion concentrations were measured at a single time point (time of resection) per subject and are represented by circles of different colors that correspond to their individual subject plasma line. Closed circles show patients assumed to be at steady state, and open circles show patients receiving the drug for the first time. Some subjects received LZD and CFZ in their background regimen, and their last CFZ and LZD doses were administered on the day preceding the resection, leading to tissue samples after 24 h. None of the patients receiving LZD and CFZ had fungal ball lesions, and those patients on CFZ also didn’t have closed nodule caseum lesions. CFZ, clofazimine; INH, isoniazid; KAN, kanamycin; LZD, linezolid; MFX, moxifloxacin; NA, not available; PZA, pyrazinamide; RIF, rifampicin.

Fig 2. Two-dimensional imaging of LZD and CFZ by MALDI-MS.

MALDI-MS ion maps of LZD and CFZ that provide a semiquantitative measure using the relative ion abundance of specific analytes in regions of interest. The highest signal intensity is fixed to 100% for each drug (red) and 0% indicating no detectable drug present (blue) on the rainbow color scale bar to the right. The caseum area is demarcated by the white line in the MALDI-MS image and corresponds to the black line of the histologically stained lesion images of the same sample in the bottom panel. A. The top panel of a nodule caseum and cavity lesion resected 20 h after LZD administration. LZD partitions favorably into cavity and nodule caseum in vivo, as indicated by the green color present in caseum and surrounding tissue showing similar concentrations in both. B. Clofazimine at steady state and 5 h post dose shows low penetration into the large necrotic granuloma in the top panel, as indicated by the blue area. CFZ, clofazimine; LZD, linezolid; MALDI-MS, matrix-assisted laser desorption-ionization mass spectrometry.

Fig 3. PK structural model.

First-line drugs R, H, and PZA and second-line drugs M, C, K, and L were modeled individually with parameters CL, V, and ka from an A describing Cps. R required a transit model to capture absorption, and H required an additional compartment (P) with Q and lag in time before absorption (Tlag) to model plasma data. Tissue concentration–time profiles were modeled with the addition of 2 parameters to describe the rate of drug absorption into the tissue compartment and the ratio of observed tissue concentration to plasma concentration. A, absorption compartment; C, clofazimine; CL, clearance; Cp, plasma concentration; H, isoniazid; K, kanamycin; ka, rate of absorption; L, linezolid; M, moxifloxacin; P, peripheral compartment; PK, pharmacokinetic; PZA, pyrazinamide; Q, intercompartmental clearance; R, rifampicin; V, volume.

Fig 4. Visual predictive check of individual lesions and drugs.

The visual predictive checks were simulated for 1,000 patients and are shown for all 7 drugs and 9 lesions. The simulated concentration–time profiles are represented by a solid black line to represent the median (typical patient) at steady state, with the shaded light blue area representing the 95% prediction interval of the 1,000 patients simulated. The red line represents the median of plasma–time concentrations after the first dose of each respective drug. The dashed black line represents the respective MIC of each drug and is displayed below the drug name. The original observed data is overlaid on top of the simulated bands as open circles to assess the ability of the model to capture both the median and distribution of the clinical data. Open red circles represent drug concentrations of patients receiving their first doses, and black open circles represent observed data for patients who were at steady state at the time of resection. First dose and steady state are overlaid in a single panel. A time-extended plasma visual predictive check is provided in S3 Fig. *PZA MIC of 12.5 mg/L is shown and is specific to an acidic environment less than pH 5.8. MIC, minimum inhibitory concentration; CFZ, clofazimine; INH, isoniazid; KAN, kanamycin; LZD, linezolid; MFX, moxifloxacin; NA, not available; PZA, pyrazinamide; RIF, rifampicin.

Fig 5. PK profiles of individual drugs in at-risk patients relative to MIC.

One thousand patients were simulated using the established model and the fifth percentile of lower exposure shown with facets for individual drugs to compare drug penetration into respective lesions shown by different colors. The dashed line in the concentration–time panel represents each drug’s respective MIC, and the bar graph below simplifies drug concentration above or below by black and white bars for each lesion. Additionally, the total time drug concentration is above MIC at steady state over 24 h is printed next to each respective bar. Importantly, PZA’s MIC of 12.5 mg/L is low-pH specific, which does not represent all the lesions’ pH environment. For the lesions with likely higher pH environments and assumed less PZA activity, gray bars instead of black bars are used to indicate drug concentration above MIC. MIC, minimum inhibitory concentration; PK, pharmacokinetic; CFZ, clofazimine; INH, isoniazid; KAN, kanamycin; LZD, linezolid; MFX, moxifloxacin; NA, not available; PZA, pyrazinamide; RIF, rifampicin.

Fig 6. Distribution of PK/PD target attainment per drug.

Exposure of 1,000 simulated patients are represented as box plots in each lesion to visualize the distributed PK/PD efficacy for individual drugs and lesions. Each panel represents an individual drug given at their standard dose until steady state is reached. The dashed lines indicate published target attainment PK/PD ratios, for which black dashed lines indicate the critical exposure values able to achieve ≥90% of maximal kill (EC90) values and gray dashed lines for drugs with published critical exposure values able to achieve ≥50% maximal kill (EC50). The percentage of the 1,000 simulated patients are printed below each box plot. Lesions are colored by their vascularity. CC, caseum from cavity; CN, closed nodule caseum; CW, cavity wall; EC90, 90% maximal effective concentrations; EC50, 50% maximal effective concentrations; FB, fungal ball; FN, caseous fibrotic nodule; FT, fibrotic tissue; Lu, lung; NN, necrotic nodule; PK/PD, pharmacokinetic/pharmacodynamic; Pl, plasma; SN, small cellular nodule.

Fig 7. Number of drugs above MIC by lesion type.

Number of drugs above MIC over time plot of first-line drugs at steady state, collected as the fifth percentile (low-exposure or at-risk patients) of 1,000 simulated patients over the 24-h dosing interval. Each square represents 1 h that a drug is above the MIC. The drugs are stacked for each hour with different colors representing different drugs. Empty squares show no drug on board. MIC, minimum inhibitory concentration.

Fig 8. Prediction of drug efficacy by lesion type.

Using defined critical PK/PD exposure values, a population of 1,000 simulated patients were defined as reaching efficacy or not from their PK simulations. The percentage of those reaching efficacy are shown as percentage box plots to compare the drugs that are able to reach a higher ratio of efficacy in specific lesions. Black boxes represent the percentage population reaching EC90, and if available, gray boxes represent the percentage population able to reach EC50. EC90, 90% maximal effective concentrations; EC50, 50% maximal effective concentrations; PK/PD, pharmacokinetic/pharmacodynamics.

Fig 9. Predicted contribution of drugs by lesion type.

A. RIF, MFX, and LZD simulations were performed at higher doses to observe their impact in hard-to-treat caseum from closed nodule. Their respective exposure distribution in 1,000 patients are shown in relation to their critical exposure value to reach EC90 (black lines), and EC50 (gray bars) if available. B. The percentage PK/PD efficacy target reached by a 1,000 simulated patient PK profiles as defined by their drug exposure versus drug MIC is shown in blue. Red boxes use CasMBC90 instead of MIC in the same set of simulated patients to compare that efficacy is both lesion and drug specific. EC90, 90% maximal effective concentrations; LZD, linezolid; CasMBC90, caseum-specific minimum bactericidal concentration; MIC, minimum inhibitory concentration; MFX, moxifloxacin; PK/PD, pharmacokinetic/pharmacodynamic; RIF, rifampicin.

Fig 10. Interactive web application.

A page of the web app is shown that displays 2 scenarios of a patient presenting with or without cavity. The user can investigate the patient’s exposure distribution, dosing, and microbiology properties.