Pharmacokinetic/Pharmacodynamic Evaluation of the Dipeptidyl Peptidase 1 Inhibitor Brensocatib for Non-cystic Fibrosis Bronchiectasis

The PPK data set comprised brensocatib concentrations pooled from healthy participants in a phase I bridging study and patients with non-cystic fibrosis bronchiectasis in the phase II WILLOW study (ClinicalTrials.gov: NCT03218917; EudraCT: 2017-002533-32) [11]. A total of 1284 steady-state brensocatib concentrations were pooled from 225 individuals who participated in either study; 68 participants with intensive plasma data in the phase I study and 157 patients from the phase II study (20 with intensive PK data and 137 with sparse PK data) were included in the PPK data set.

The phase I study was a single- and repeat-dosing study conducted in two parts. Part A was a randomized, double-blind, placebo-controlled, dose-escalation phase in which healthy Japanese and White adults were randomized 4:1 to receive oral doses of brensocatib 10, 25, or 40 mg or placebo after an overnight fast of ≥ 10 h (n = 6 per dose per race); participants received a single dose on day 1 followed by daily doses on days 4–30 [12]. Blood samples for PK analysis were collected pre-dose on days 1 and 30 and post-dose at 0.5, 0.75, 1, 1.5, 2, 3, 4, 8, 12, 24, 48, and 72 h. An additional sample was drawn pre-dose on day 14. Part B was an open-label, two-period, two-sequence crossover phase in which newly enrolled healthy Japanese and White participants (n = 10 per race) were randomized 1:1 to receive single doses of brensocatib 40 mg under either fasted or fed (high-fat, high-calorie breakfast) conditions on days 1 and 8 [13]. Blood samples for PK analysis were drawn pre-dose on days 1 and 8 and post-dose at 0.25, 0.5, 1, 1.5, 2, 3, 3.5, 4, 6, 8, 10, 12, 24, 36, and 72 h. Participants in both parts A and B were nonsmoking adults (18–50 years of age) with no notable abnormalities on a 12-lead electrocardiogram, a body mass index of 18.0–30.0 (calculated as weight in kilograms divided by height in meters squared), and a body weight of ≥ 45–< 100 kg. Except for paracetamol/acetaminophen or ibuprofen, use of over-the-counter medications (including vitamins) 7 days before screening, or prescription medications and herbal remedies 30 days before screening, was not permitted until the end of treatment.

The phase II study was a randomized, double-blind, placebo-controlled trial in which adults (18–85 years of age) with computed tomography-confirmed non-cystic fibrosis bronchiectasis and two or more documented exacerbations in the previous 12 months were randomized 1:1:1 to receive brensocatib 10 mg, brensocatib 25 mg, or placebo once daily (QD) for 24 weeks. Intensive blood samples for PK analysis were collected from 20 active participants pre-dose on day 1 and at week 4 and post-dose at 1, 2, 3, 4, 6, and 8 h. Trough samples were to be collected pre-dose at weeks 2, 12, 24, and 28 from 138 participants. Detailed methods and results from the WILLOW study were previously published [11].

Brensocatib was extracted from plasma samples treated with an anticoagulant (di-potassium ethylenediaminetetraacetic acid) and quantified using a validated method of tandem liquid chromatography–mass spectrometry with a quantitation range of 0.25–150 ng/mL.

All data management activities were performed using SAS version 9.4 software. The PPK analysis was conducted using NONMEM version 7.4.3 (ICON Development Solutions, Ellicott City, MD, USA), implementing first-order conditional estimation with η–ε interaction, as facilitated by Perl-Speaks-NONMEM version 4.7.0 and Pirana version 2.9.9 (Certara). Covariate analysis was conducted using the automated stepwise covariate modeling algorithm implemented in Perl-Speaks-NONMEM. Creatinine clearance (CLcr) was calculated from baseline serum creatinine, age, and weight (kg) using the Cockcroft–Gault equation [14] and then normalized to body surface area (BSA). All other analyses, including calculation of post-hoc PK parameters and Monte Carlo simulations, were performed using the R statistical software package version 3.6.

Demographic information (age, weight, height, BSA, and sex) and clinical characteristics (serum albumin and serum creatinine) were used to characterize the analysis population and evaluate the interindividual variability for selected PK parameters.

Because of the difficulties associated with fitting a model to the pooled first-dose and steady-state data across the two studies, first-dose PK data were excluded. Brensocatib plasma concentrations were also excluded from the PPK analyses if the dates and times of the previous dose or sample collection or any covariate information (demographics or laboratory data) were missing. Concentrations below the lower limit of quantification (LLOQ) were set to missing.

Models were selected based on a comparison of the minimum value of objective function (MVOF) for nested models, evaluation of individual- and population-predicted accuracy, graphic examination of goodness-of-fit plots, reduction in both intraindividual variability (σ²) and interindividual variability (ω²), Akaike’s information criterion and/or the Bayesian information criterion for non-nested models [15], and biological plausibility. Throughout model development, normalized prediction distribution errors (NPDEs) were computed as an additional model diagnostic to assess bias, with a resemblance to a standard normal distribution [N(0, 1)] expected for a model without bias. The condition number (derived as the ratio of largest to smallest eigenvalue) was also assessed for possible overparameterization; a model with a condition number < 1000 was considered not overparameterized [16, 17].

A base structural model was developed using PK data from the healthy participants to evaluate the number of distributional compartments (1–3), oral absorption kinetics, and error models. Once an adequate model was developed, PK data were pooled with the data from patients with non-cystic fibrosis bronchiectasis in the phase II study to test the adequacy of the structural model. Following confirmation of the structural model, a formal covariate analysis was conducted using stepwise forward inclusion (p < 0.05; ΔMVOF ≥ 3.84 U) and backward elimination (p < 0.001; ΔMVOF ≥ 10.8 U). Covariates evaluated include age, weight, height, BSA, sex, race, serum albumin, and CLcr. The final PPK model was evaluated using sampling-importance resampling to assess the uncertainty distribution of each of the predicted parameters [18, 19]. Prediction-corrected visual predictive checks (pc-VPCs) were used to determine the accuracy of model-predicted concentration values [20].

The final PPK model was used to predict the individual concentration profiles and PK parameters—including apparent oral clearance (CL/F), apparent oral volume of distribution of the central compartment (Vc/F), terminal elimination half-life (t_½), maximum plasma concentration (C_max), and area under the plasma concentration–time curve over the dosing interval (AUC_τ)—in both healthy participants and patients with non-cystic fibrosis bronchiectasis.

Pulmonary exacerbations were defined as the presence of three or more of the following symptoms for 48 h that resulted in a physician’s decision to prescribe antibiotic treatment: hemoptysis, fatigue, malaise, change in sputum consistency, decreased exercise tolerance, and increased cough, sputum volume, sputum purulence, or breathlessness [21]. Sputum samples for determination of neutrophil elastase activity were collected at screening (days −42 to −1); baseline (day 1); weeks 2, 4, 12, and 24; and follow-up (week 28). Neutrophil elastase activity was measured by ProteaseTag Active Neutrophil Elastase Immunoassay (ProAxsis Ltd, Belfast, UK). Safety data included adverse events of special interest (AESIs)—such as periodontal disease, hyperkeratosis, and infections other than pulmonary infections—that were selected based on the mechanism of action of brensocatib.

Exploratory analyses were conducted to determine any potential PK/PD relationships between the independent PK exposure measures (AUC_τ, C_max, and minimum plasma concentration [C_min]), derived from the fit of the PPK model, and the dependent variables (pulmonary exacerbations, sputum neutrophil elastase, and AESIs). Age (< 65 and ≥ 65 years; < 75 and ≥ 75 years), baseline sputum neutrophil elastase concentration below the limit of quantification (BLQ), use of rescue medications, use of macrolides for prevention of pulmonary exacerbations, and sputum culture positive for Pseudomonas aeruginosa were identified as potential covariates to ensure that any PK/PD relationships associated with brensocatib exposure were independent of potential predictors. Exacerbations were evaluated as dichotomous (yes or no for occurrence within the 24-week treatment period) and by using time-to-event models for time to occurrence within 24 weeks.

Sputum neutrophil elastase was evaluated as both a dichotomous variable (for attaining one or more post-baseline concentration below the LLOQ) and a continuous variable (absolute values or change over time and maximum change from baseline). When sputum neutrophil elastase was evaluated as a continuous variable, BLQ values were imputed as one-half of the reported LLOQ for that observation (LLOQ could vary among samples based on the extent of dilution necessary for that sample). When sputum neutrophil elastase was evaluated as a categorical variable, values below the LLOQ were retained in the data set. Post-baseline values below the LLOQ that were achieved after the occurrence of a pulmonary exacerbation were excluded because the administration of antibiotics for an exacerbation has the potential to suppress sputum neutrophil elastase activity [22].

Brensocatib exposures were evaluated as continuous variables in exposure quartiles; patients who received placebo were assigned a quartile of 0. Results from the exploratory PK/PD analyses were graphically displayed to examine potential relationships between exposure and the dependent variables. Continuous independent variables were evaluated both as continuous and categorical independent variables. Categorical transformations of the independent variables were made based on a recursive partitioning algorithm designed to identify thresholds above and below which substantial differences in the dependent variables were apparent. For this analysis, recursive partitioning was accomplished using classification-and-regression tree (CART) algorithms as implemented in the R statistical software package. If potential PK/PD relationships were identified for the dichotomous endpoints, univariate evaluations were conducted using contingency tables (using χ² or Fisher exact test) for the dichotomous independent variables, and logistic regression was used for continuous independent variables. Univariable relationships for continuous endpoints (e.g., maximum change in sputum neutrophil elastase) were evaluated using correlation coefficients (Pearson χ² and Spearman rank correlation coefficient) and linear regression. Relationships for time-to-event endpoints (e.g., occurrence of pulmonary exacerbation) were examined using log-rank tests for categorical independent variables and Cox proportional hazards regression for continuous independent variables.

Clinical outcomes, including inhibition of sputum neutrophil elastase and reduction in the occurrence of pulmonary exacerbations, were simulated based on the PK–efficacy relationships to support the selection of brensocatib doses for future clinical studies. A population of 1000 patients with non-cystic fibrosis bronchiectasis was generated by resampling body weight and CLcr from the phase II study population (n = 255). Each patient was assigned a value for CL/F based on the PPK model using the resampled median for the CLcr effect on CL/F and the estimated interindividual variability (IIV) in CL/F from the final population PK model per the following equation:

Steady-state drug exposure within a dosing interval (AUC_τ) was then predicted from CL/F for each simulated patient for brensocatib doses of 2.5, 5, 10, 25, and 40 mg QD according to the following equation:

A bootstrapping approach was used to simulate likely outcomes in terms of sputum neutrophil elastase and pulmonary exacerbation by replicating the simulation process 500 times. For each replicate, the probability of achieving post-baseline sputum neutrophil elastase BLQ and the probability of the occurrence of a pulmonary exacerbation was recalculated by sampling the observed data with replacement. In this manner, it was possible to obtain a distribution of predicted outcomes at each brensocatib dose that more robustly accounted for the variability in the observed outcomes and the relatively small sample sizes.

Participants in the phase I study were predominantly male, with a mean age of 36.2 years; the number of Japanese and White participants was equal (Table 1). Patients with non-cystic fibrosis bronchiectasis from the phase II WILLOW study were mostly female and White, with a mean age of 64.1 years (Table 1). Other than the slightly lower brensocatib concentrations observed in the 10-mg group in the phase I study, dose-normalized brensocatib plasma concentrations were consistent across doses and between studies (Fig. 1).

Using data from the healthy participants in the phase I study, several potential models to describe brensocatib absorption and elimination kinetics were attempted. For the absorption model, simple first-order models and more complex transit-chain models were attempted. For elimination, 1-, 2-, and 3-compartment models with linear elimination were attempted; nonlinear elimination models were not explored because they were not applicable based on the observed data. Brensocatib PK profiles were best described with two disposition compartments and a chain of six transit compartments with separate absorption rate constants (k_tr) for fed and fasted states (Supplementary Fig. 1). This base model was then applied to the pooled PPK data set, and goodness-of-fit plots indicated that the model provided a robust fit to the observed concentrations, with minimal bias based on the distribution of NPDEs. PPK parameters for the model were well estimated, with acceptable interindividual and residual variabilities. The condition number was low (23.3), indicating that the model was not overparameterized.

Using the forward inclusion/backward elimination procedure described in the methods, two statistically significant covariate relationships were identified: CLcr on CL/F (CL/F × [CLcr/75.2]^0.276) and age on Vc/F (Vc/F × [age/60]^0.396). Since all parameters were dependent on body weight using allometric principles [23], the additional covariate relationships were significant after accounting for the effect of body weight on PK. Other covariates tested (sex, race, body weight, and serum albumin levels) did not have a significant relationship with the variability in PPK model parameters. PPK parameters from the fit of the final model, along with the summary statistics from the resampled parameters, are provided in Table 2. Goodness-of-fit plots for the final model indicated a lack of systematic bias and good precision, with r² = 0.62 for population predictions versus observations and r² = 0.96 for individual predictions versus observations (Supplementary Fig. 2, see ESM). The NPDE plots also did not indicate bias (Supplementary Fig. 3, see ESM). Qualification of the final model using simulation-based diagnostics (pc-VPC) plots indicate that the final model adequately described both the central tendency and variability in observed brensocatib concentrations over time for both the healthy participants in the phase I study and patients with non-cystic fibrosis bronchiectasis in the phase II WILLOW study (Fig. 2).

The mean estimated steady-state AUC_τ from the final PPK model was 1760 in patients treated with brensocatib 10 mg and 4540 ng·h/mL in patients treated with brensocatib 25 mg compared with 1110 and 3660 ng·h/mL, respectively, in healthy participants (Table 3). In the brensocatib 10- and 25-mg groups, the mean t_½ values were 38.5 h and 39.1 h, respectively, in patients with non-cystic fibrosis bronchiectasis compared with 24.3 h and 26.0 h in healthy participants (Table 3). The PPK-predicted parameters were comparable to those determined using a noncompartmental analysis method based on observed PK concentrations [10]. Estimates for steady-state AUC_τ, C_max, and C_min by age (Supplementary Fig. 4), CLcr (Supplementary Fig. 5), and body weight (Supplementary Fig. 6, see ESM) in patients in the phase II study were consistent across doses, indicating that dose adjustments are unlikely to be warranted in patient subpopulations.

A greater percentage of patients treated with brensocatib had sputum neutrophil elastase levels BLQ compared with patients treated with placebo (76.8% vs 50.0%). The incidence of neutrophil elastase BLQ was the greatest (86.8%) in patients whose brensocatib AUC_τ was in the highest quartile (4845–9466 ng·h/mL) compared with 77.5% in the lowest AUC_τ quartile (624–1758 ng·h/mL) and 50.0% in the placebo group (Table 4). Additionally, two AUC_τ thresholds (1150 and 3811 ng·h/mL) were identified via CART for attaining neutrophil elastase BLQ: 50% with AUC_τ <1150 ng·h/mL, 72.7% with AUC_τ 1150–3810, and 86.4% with AUC_τ > 3810 ng·h/mL.

The incidence of pulmonary exacerbations was lower in patients in the pooled brensocatib group than in patients in the placebo group (34.4% vs 50.0%), with the incidence ranging from 30.0% in those in the first brensocatib AUC_τ quartile (624–1758 ng·h/mL) to 41.0% in the third quartile (2840–4824 ng·h/mL) (Table 4). The time to pulmonary exacerbation was reduced in each AUC_τ quartile relative to placebo (Supplementary Fig. 9, see ESM); however, no clear relationship between higher exposure and longer time to pulmonary exacerbation was evident.

Exploratory analyses of sputum neutrophil elastase as a continuous variable revealed substantial within-patient variability such that the absolute values or change over time did not predict outcome. When evaluated as a dichotomous variable, sputum neutrophil elastase was found to be a predictor of pulmonary exacerbations. Patients who achieved neutrophil elastase BLQ post-baseline experienced significantly fewer exacerbations and a prolonged time to first exacerbation compared with those with persistently detectable neutrophil elastase. The hazard ratios (95% CI) were 0.32 (0.21–0.48; p < 0.0001) for all patients (Fig. 3a) and 0.29 (0.17–0.51; p < 0.0001) for brensocatib-treated patients (Fig. 3b).

The PK/PD analyses for safety did not reveal any clinically relevant relationships between brensocatib exposure and the incidence of AESIs (Fig. 4; Supplementary Table 1, see ESM). Although a possible relationship between steady-state AUC_τ and the occurrence of periodontal disease at week 8 was suggested by objective criteria, the relationship was not statistically significant.