Method
Study design
We performed a longitudinal pharmacokinetic study in pregnant women ≥ 18 years of age treated for RR-TB, and their infants, at King Dinuzulu Hospital in Durban, Kwazulu-Natal - nested within a cohort, which has been previously described. (13) KDH is a specialist provincial RR-TB hospital where, until recently, all pregnant women with RR-TB were referred for care. With some individual regimen variability, all participants were treated with a minimum of five drugs including bedaquiline. Other drugs included: pyrazinamide, isoniazid, clofazimine, linezolid, moxifloxacin, and less commonly: ethambutol, terizidone, levofloxacin, ethionamide and para-aminosalycylic acid. We performed pharmacokinetic sampling pre-dose, and two, four and six hours post-dose in the third trimester of pregnancy (≥28 weeks), and at the six week postpartum visit. Dosing on both sampling days was observed after a standard breakfast; the tablets/capsules were ingested with 250 mL of water. Considering bedaquiline is dosed three times a week (after the two week loading dose), it was not always logistically possible to schedule pharmacokinetic sampling on a day when bedaquiline was administered. We therefore recorded the last date and time when bedaquiline was dosed to interpret the exposures with our modelling. All concurrent medications, including antiretroviral therapy were recorded. If available, human milk samples were taken from breastfeeding mothers by manual expression at the same timepoints that blood was drawn at the postpartum visit (i.e. pre-dose, and two, four and six hours post-dose); samples were frozen within 30 minutes of sampling at minus 80° C. To evaluate infant drug exposure, a single random plasma sample was taken from infants at the postpartum visit. If applicable, the time of the most recent breastfeeding prior to the infant blood draw was recorded.
Pharmacokinetics
Plasma and human milk samples were stored at minus 80°C and transported to the University of Cape Town, Division of Clinical Pharmacology laboratory where total plasma and human milk bedaquiline assays were performed using liquid chromatography with tandem mass spectrometry. The plasma assay for total bedaquiline has previously been described. (14) Bedaquiline and its M2 metabolite in human milk were analysed with a validated assay developed at the Division of Clinical Pharmacology laboratory; the standards and quality checks were performed using blank donated human milk. The extraction procedure consisted of protein precipitation and solid phase extraction, followed by gradient liquid chromatography on an Agilent Poroshell 120 SB-C18 (2.1 mm x 50 mm, 2.7 μm) analytical column with tandem mass spectrometry detection. An AB Sciex API 3000 mass spectrometer at unit resolution in the multiple reaction monitoring mode was used to monitor the transitions of the protonated precursor ions m/z 555.1, m/z 561.1, m/z 541.1, and m/z 545.1 to the product ions m/z 58.2, m/z 64.1, m/z 480.3, and m/z 480.4 for bedaquiline, TMC207-d6, M2, and M2-d3C13, respectively. Electro Spray Ionisation was used for ion production. The calibration curves fitted quadratic (weighted by 1/x) regressions based on peak area ratios over the ranges 0.0780 – 5.00 µg/mL for bedaquiline and 0.0312 – 2.00 µg/mL for M2. The combined accuracy (%Nom) and precision (%CV) statistics of the lower limit of quantification, low, medium, and high quality controls of bedaquiline and M2 during intra- and inter validations were between 96.7% and 106.5%, and 3.4% and 7.5%, respectively.
Modelling
Bedaquiline concentrations were interpreted using population pharmacokinetic modelling in NONMEM version 7.4.5 (15). Perl-speaks-NONMEM version 5.2.6, Pirana version 3.0, and R with the package xpose4 were used to facilitate the model development process, data manipulation, and generation of model diagnostics (16). As a starting point, we used a published population pharmacokinetic model of bedaquiline in non-pregnant adults with HIV and drug-resistant tuberculosis (14). Briefly, the published model consists of three disposition compartments for bedaquiline and one disposition compartment for M2. There was a correlation between bedaquiline and M2 between-subject-variability on clearance, and residual variabilities. The effect of body weight on all disposition parameters was included using allometric scaling, and albumin also affects the drug disposition parameters. The co-administration of ritonavir-boosted lopinavir reduced bedaquiline and M2 clearance by 65% and 42%, respectively. Molar concentrations were used during model development to account for mass balance between bedaquiline and its metabolite M2. Participant albumin information were not captured in the current study, therefore we imputed a reported albumin concentration from a previously study in South Africa patients with RR-TB (17).
When analysing the data, we first fit the original model as published, without re-estimating any of the population parameters, but using the study covariate, doses and dosing regimen information. This is similar to using the current data as an “external” validation of the model, i.e. assessing how the previous model predicts the current data based solely on covariate information and assuming no effect of pregnancy (which was not part of the origina model). Afterwards, we attempted to use the data to re-estimate parameter values, using the general principles of model development including drops in NONMEM objective function value (OFV) for assessment of statistical significance and inspection of diagnostic plots.

Calculation of infant bedaquiline intake with human milk

To estimate how much bedaquiline is ingested per day by a typical child breastfed by a mother receiving bedaquiline, we assumed an average infant milk ingestion of 0.15 L/kg/day (18). The following equation was used to calculate the infant dose (19):
\begin{equation} D_{\text{infant}}=\ C_{m}\bullet\ V_{m}\nonumber \\ \end{equation}
where Vm is the volume of milk ingested by breastfeeding, andCm is the drug concentration in human milk. This was calculated using the formula below:
\begin{equation} C_{m}=M:P\ \bullet\ C_{\text{pss}}\nonumber \\ \end{equation}
where Cpss is the average maternal plasma level at a steady-state, and M:P is the human milk-to-plasma ratio.
Ethics
Ethics approval for the study was granted by the South African Medical Research Council Ethics Committee (EC017-6/2016), and the University of Cape Town Human Research Ethics Committee (HREC: 666/2018). Informed consent was taken from all participants in a language of their choice (either English or isiXhosa).