Mechanistic Physiologically Based Pharmacokinetic Modeling of Dry Powder and Nebulized Formulations of Orally Inhaled TMEM16A Potentiator GDC‐6988

ABSTRACT

The orally inhaled route of administration for respiratory indications can maximize drug exposure to the site of action (lung) to increase efficacy while minimizing systemic exposure to achieve an improved safety profile. However, due to the difficulty of taking samples from different regions of the human lung, often only systemic pharmacokinetic (PK) samples are taken and assumed to be reflective of the lung PK of the compound, which may not always be the case. In this study, a mechanistic lung physiologically based pharmacokinetic (PBPK) model was built using a middle‐out approach (i.e., combining elements of bottom‐up prediction and using clinical data to inform some model parameters) to predict plasma and lung PK of an orally inhaled TMEM16A potentiator GDC‐6988 in humans. The lung PBPK model accounted for lung deposition, lung and oral absorption, systemic clearance, and tissue distribution. The model was refined using data from a Phase 1b study with dry powder (DP) formulation and was also verified using data from a Phase 1 study with a nebulized (Neb) formulation. The refined model adequately captures the observed GDC‐6988 plasma PK profiles in both the DP and Neb studies and allows prediction of the regional lung fluid and tissue concentrations. The sensitivity analyses showed that the systemic C _max depended on the ratio of airway to alveolar deposition, but this did not impact the AUC. This novel mechanistic lung PBPK modeling framework could be applied to predict plasma and regional lung exposure and inform the early clinical development of inhaled molecules (e.g., dose selection).

Summary.

• What is the current knowledge on the topic?

  • ○
    Orally inhaled route of administration with direct drug delivery to the site of action is often considered for molecules intended to treat respiratory diseases.
• What question did this study address?

  • ○
    Due to the difficulties in taking PK samples at different regions of the human lung, one key clinical pharmacology challenge for orally inhaled molecule development is that the observed systemic exposure may not fully reflect the drug exposure in the lung (site of action). The traditional exposure‐response analysis using systemic exposure may be less useful in supporting the dose selection for inhaled molecules. Therefore, lung PK prediction via a modeling and simulation approach based on data from in vitro and preclinical experiments and clinical studies can be valuable to support the clinical development of inhaled molecules.
• What does this study add to our knowledge?

  • ○
    A mechanistic lung PBPK modeling framework that captures lung deposition and absorption, intestinal absorption, tissue distribution, and systemic clearance was established and used to predict systemic and lung exposure for an orally inhaled small molecule GDC‐6988 under clinical development for muco‐obstructive lung disease.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    With this novel mechanistic lung PBPK model framework, we could predict the systemic and lung PK and inform the clinical development of orally inhaled molecules (e.g., dose selection).

1. Introduction

Using the orally inhaled route of administration to treat respiratory diseases can provide numerous benefits for patients. The direct drug delivery to the site of action (lung) can result in higher lung drug exposure and retention than dosing by the oral route, and subsequently increase the opportunity of achieving a desired clinical outcome. Given that inhaled molecules are generally designed to have low intestinal bioavailability and high systemic clearance, despite part of the delivered dose being trapped in the mouth/throat area and eventually swallowed, the systemic drug exposure achieved via the intestine is often low, contributing to an enhanced safety profile [1]. In addition, orally inhaled drugs also provide patient convenience compared to injectables, especially for patients afraid of needles.

Although it is believed that lung concentration is the key driver for efficacy and local safety events, one challenge in inhaled drug development is that generally only systemic pharmacokinetics (PK) samples are taken and measured in clinical studies due to the difficulty in obtaining PK samples from different regions in the human lung. Despite systemic drugs primarily being absorbed through the lung, it is not clear how well PK profiles observed in the systemic circulation reflect those in the lung, especially in the different regions of the lung. Therefore, estimation of the lung PK via combinations of novel approaches (e.g., in silico, imaging, PK studies) is of great interest in inhaled drug development [2, 3, 4, 5, 6, 7, 8, 9]. For example, observed systemic PK from the inhaled route (and intravenous route) coupled with in vitro characterization of the formulations and empirical PK modeling is the widely applicable and practical approach available for predicting lung mass profiles and distinguishing the performance characteristics of the orally inhaled drug products [2, 3, 4, 5]. Other approaches such as imaging and pharmacodynamic endpoint bioequivalence may be applicable on a case‐by‐case basis [6, 7, 8]. PK studies can also be used to provide relevant information on the pulmonary performance characteristics (i.e., available dose, residence time, and regional lung deposition) to inform lung PK prediction [9].

Extrapolating in vitro data to predict in vivo drug absorption, distribution, metabolism, and excretion via a physiologically‐based pharmacokinetic (PBPK) model is a common approach for oral and injectable drugs [10], but it has been relatively infrequently applied to inhaled drugs. Despite the challenges in collating the appropriate physiology and physical/chemical data needed to parameterize lung deposition of inhaled compounds, mechanistic models describing the lung deposition, absorption, and/or distribution of drugs into different regions of the lung tissue have emerged in the last decade [5, 11, 12, 13]. Data from lung PBPK modeling may in turn allow the calculation of more meaningful exposure‐response analysis (e.g., with predicted lung exposure) to support dose selection in future clinical trials. Due to a lack of human lung tissue samples/data, direct verification of the PBPK modeling for predictions of lung tissue concentration is often not possible without additional experiments (either in pre‐clinical species or in humans).

GDC‐6988 is an inhaled, selective, and potent potentiator of transmembrane member 16A (TMEM16A), a calcium‐activated chloride channel that is expressed in the airway epithelium. GDC‐6988 is under investigation as a potential treatment of muco‐obstructive lung diseases (e.g., cystic fibrosis, chronic obstructive pulmonary disease, non‐cystic fibrosis bronchiectasis). Patients with muco‐obstructive lung diseases commonly have elevated mucin concentrations, diminished mucociliary clearance, and subsequent inflammation, infection, and obstruction in the airways [14, 15, 16]. TMEM16A potentiation with inhaled GDC‐6988 has the potential to restore airway mucus hydration and provide clinical benefit through improved mucociliary clearance. Inhaled GDC‐6988 administered via nebulized suspension (Neb) and dry powder inhalation (DPI) has been studied in two Phase 1 studies in healthy subjects [17, 18].

The aim of this study was to build a mechanistic lung PBPK model with a combined bottom‐up and top‐down approach (or middle‐out approach) to describe the observed GDC‐6988 plasma PK with DPI administration, and then verify the model with observed plasma data with Neb administration. The ultimate goal of this work was to use the mechanistic lung PBPK model for the prediction of the systemic and regional lung exposure of GDC‐6988 and in turn to use this information to support dose selection for further trials of the compound. Regional lung deposited dose was extrapolated from particle size and delivered dose measurements using a novel adaptation of a particle lung deposition model [19]. Sensitivity analyses were conducted for key parameters affecting deposition, dissolution, lung transport, and systemic clearance. The lessons learned from this exercise and current model limitations are also discussed.

2. Materials and Methods

2.1. Clinical Studies

2.1.1. Phase 1 Neb Study

Study ET‐TMEM‐01 was a first‐in‐human, randomized, double‐blind, placebo‐controlled interventional study to assess the safety, tolerability, and PK of single and multiple ascending doses of inhaled nebulized GDC‐6988 (ETD002) with PARI LC Sprint nebulizer in healthy volunteers (NCT04488705) [17]. The study evaluated single doses of GDC‐6988 up to 150 mg and repeated doses up to 75 mg twice daily (BID) for up to 14 days. The dosing information (e.g., nominal dose, emitted dose, and nebulizing duration) of this study is summarized in Table S1.

2.1.2. Phase 1b DPI Study

After the Phase 1 Neb study, a dry powder (DP) formulation of GDC‐6988 was developed for use in future clinical studies. The Phase 1b randomized, double‐blind, placebo‐controlled study (GB43838, ISRCTN30841680) assessed the safety, tolerability, and PK of multiple ascending doses of inhaled (with a customized RS01 Plastiape device) DP GDC‐6988 with and without albuterol pretreatment in healthy volunteers. Subjects received a nominal dose of 11.2, 28, and 42 mg BID GDC‐6988 or placebo for 14 days, with albuterol pretreatment starting on Day 8 for the first two cohorts and Day 1 for the 42 mg BID cohort [18]. Additional dosing information (e.g., nominal dose, number of capsules, and total delivered dose) of this study is summarized in Table S2.

Both the Phase 1 Neb and Phase 1b DPI studies were approved by the institutional review board or independent ethics committee and were conducted in accordance with the Declaration of Helsinki and Good Clinical Practice Guidelines. Training on the use of the device was provided for all subjects in both studies to ensure consistency in the administration of the inhaled dose.

2.2. Mechanistic Lung PBPK Model Development

The Simcyp population‐based PBPK Simulator (Simcyp version 23, Certara UK) was used to develop the mechanistic lung PBPK model. The structure and physiology of the permeability‐limited lung model used in this work were described previously [13]. The model describes the lung with seven regions: two airway regions, Upper Airway (UA) and Lower Airway (LA) and five Alveolar‐Interstitial (AI) regions, Right Top (RT), Right Middle (RM), Right Lower (RL), Left Top (LT) and Left Middle (LM). Each region is composed of three compartments: lung lining fluid, tissue mass, and blood. After inhalation, solid particles are deposited in the fluid compartments where dissolution occurs. Macrophage and mucociliary clearance transport drugs between the fluid compartments of the regions toward the upper airway. Deposition in different regions of the lung was predicted using information on the physical properties of the inhaled drug formulation and the ICRP66 model [19].

A combined bottom‐up and top‐down approach (or middle‐out approach) was used to build the mechanistic lung PBPK model [20]. The key components of the mechanistic lung PBPK model include lung deposition, lung absorption, intestinal absorption, tissue distribution, and systemic clearance. Figure 1 illustrates the lung deposition and lung absorption model in detail. Initially, human in vitro data were used to parameterize the GDC‐6988 PBPK model (i.e., systemic clearance (CL) and volume of distribution (Vss)). Key parameters from animal studies with oral and intravenous (IV) administrations are summarized in Table S3. Modeling of preclinical species indicated that in vitro hepatocyte clearance was a poor predictor of in vivo CL, and Vss was overpredicted by physicochemical‐based predictors. Therefore, the GDC‐6988 human systemic CL and Vss initially predicted based on in vitro data were further optimized/refined with the observed plasma PK data from the Phase 1b DPI study. Finally, the refined model was verified with the observed PK data from the Phase 1 Neb study. Key input parameters used in the model development are summarized in Table 1.

FIGURE 1.

A schematic of the Simcyp lung module with lung deposition and lung absorption. Each of the seven regions (five for lung alveolar‐interstitial regions and two for airway) is split into three compartments, blood (red), tissue mass (blue) and lining fluid (gray). Solid particles deposited in the fluid regions undergo dissolution and mucociliary clearance. Inhaled dose is assumed to deposit to four main deposition regions (i.e., extrathoracic (ET), bronchial (BB), bronchiolar (bb), and alveolar‐interstitial (AI)) with the AI deposition determining the dosing into the five lung regions (3 regions in the right lung and 2 regions in the left lung).

TABLE 1.

Input parameters and dosing information used for human PBPK models of GDC‐6988.

Parameter	Value	Method/reference
Molecular weight (g/mol)	361.82
Log P	4.29	Measured
Compound Type	Neutral
B:P	1	Measured
fuplasma	0.0163	Measured
Main plasma binding protein	Albumin	Assumed as compound neutral
Intestinal absorption model	ADAM
Permeability	Predicted from MDCK II
P app (cm/s)	8.99 × 10−6	MDCK II
P app Scalar S Papp	8.8411	See Methods
P eff,man (cm/s)	10.245 × 10−4	See Methods
Aqueous solubility (mg/mL)	0.094
fugut	1	Default
Distribution model	Full PBPK model
Tissue partition coefficients	Method 2	[21]
K p Scalar	0.11	Manual refinement with DPI Phase 1 plasma PK data
Lung model
Permeability (cm/s)	0.97478 × 10−4	Predicted from log D6.5 [13]
fumass	0.006	Measured
fufluid	0.170	See Methods and Results
DPI dissolution model parameters
Lung fluid solubility (mg/mL)	0.038	Measured in a simulated lung lining fluid Survanta
DPI deposition parameters
Particle mean radius (μm)	1.38	See Section S.1: Data S1
Particle radius CV (%)	50.7	See Section S.1: Data S1
Particle density (g/mL)	1.2	Default
Shape factor χ	1.5	Default
Neb deposition parameters
Activity	Sitting
Fraction nose (F n)	0.7	Default [19]
Particle mean radius (μm)	4.2	See Section S.1: Data S1
Particle radius CV(%)	71.4	See Section S.1: Data S1
Particle density (g/mL)	1.2	Default
Shape factor χ	1.5	Default
Elimination
UGT2B7 CLint (μL/min/mg microsomal protein)	1400	See Methods
UGT1A9 CLint (μL/min/mg microsomal protein)	1350	See Methods
CYP2D6 CLint (μL/min/pmol isoform)	120	See Methods
CYP2C9 CLint (μL/min/pmol isoform)	9	See Methods
CYP3A4 CLint (μL/min/pmol isoform)	4.5	See Methods
Additional clearance (L/h)	194	Manual refinement with DPI Phase 1 plasma PK data

Note: Other model parameters not stated are kept as default.

2.2.1. Lung Deposition

The deposition of inhaled GDC‐6988 as Neb or DP formulations was predicted using the ICRP66 lung deposition model [19]. The model splits the lung into four regions: the extrathoracic (ET), bronchial (BB), bronchiolar (bb), and alveolar‐interstitial (AI) (Figure 1). The deposition efficiencies for inhalation and exhalation are computed (using the aerodynamic and thermodynamic diameters of the inhaled particles) and summed to give deposition in each of the four regions, and the remaining dose fraction is assumed to be exhaled. The fraction deposited in the ET region is the fraction swallowed, while the fraction in the three other lung regions is the fraction inhaled and deposited to the airways and the lung. Parameters describing the air movement during breathing include the fraction inhaled through the nose (f _n), tidal volume (V _T), volumetric flow rate (V̇), and functional residual capacity (FRC). The physiological parameters within a population vary with the sex of the individual and were estimated using the data described in ICRP66. Deposition can be computed for a distribution of particle sizes described by the mean geometric radius and coefficient of variation (CV). The mass median aerodynamic diameter (MMAD) and geometric standard deviation (GSD) from in vitro next generation impactor (NGI) studies (Table S4) were converted to mean geometric radius and CV by the process described in the Section S.1: Data S1. To simulate the nebulized dose, the total nebulized dose was split into 10 doses administered at equal intervals within the nebulizing duration to approximate inhalation of the nebulizer dose over the course of the estimated administration time (up to 23 min).

The default ICRP66 model used for the nebulizer administration was adapted to account for the deposition that occurs in subjects who hold their breath after the administration of drugs via a dry powder inhaler [22]. A detailed description of the adapted model is provided in the Section S.2: Data S1 and Tables S5 and S6.

2.2.2. Lung Absorption

Solid particles of the emitted dose are assumed to be immersed in the epithelial lining fluid of the lung at the moment of administration. Solid particles and dissolved compounds are subject to fluid transport including mucociliary clearance. This linear process acts to move particles and compounds from alveolar to airway and extrathoracic regions. Clearance rates were taken from the ICRP 66 report [19] and are given in the Section S.3: Data S1. The dissolution rate of the particles is given by the Wang and Flanagan equation [23, 24]. The aqueous diffusion coefficient (D) is predicted from the compound's molecular weight, and the effective diffusion layer thickness (h _eff) is calculated using the Hintz‐Johnson relationship [25]. Lung lining fluid solubility was measured in an artificial lung lining fluid (Survanta, containing a modified bovine pulmonary surfactant). The fraction unbound of GDC‐6988 in epithelial lining fluid (fu_fluid) was estimated using the measured fu in plasma (0.0163) correcting for the difference in albumin content between alveolar epithelial lining fluid (3.7 mg/mL) [26] and plasma, assuming that the affinity of GDC‐6988 to albumin is the same in both matrices.

Transport of the compound across the epithelial lining fluid, lung mass, and alveolar blood compartments is calculated as described previously using log D pH 6.5 and the number of hydrogen bond donors as model inputs [13]. Apical and basal membrane permeability was set to be identical for all regions. Binding of GDC‐6988 to lung tissue homogenate was measured by equilibrium dialysis.

To simulate the deposition of Neb aerosol particles, the dissolution model parameters were adjusted so that the total dissolution of particles occurs within the first few seconds after the administration. This was done by setting the lung fluid solubility to 1 mg/mL and increasing the diffusion layer model scalar (S _DLM) from a default value of 1–10. This approximates the dosing of the dissolved compound directly to lung fluid, assuming that the nebulized suspension inhaled makes a negligible difference to lung fluid volume.

2.2.3. Intestinal Absorption

The intestinal absorption of GDC‐6988 was simulated using the multi‐compartment Advanced Dissolution, Absorption and Metabolism (ADAM) model [27]. The fraction of the dose deposited in the ET region following inhalation is transferred to the stomach compartment of the intestinal model at the moment of administration. The compound cleared by mucociliary clearance in the upper airway is assumed lost rather than swallowed. For GDC‐6988, the impact of this is negligible due to its low predicted oral bioavailability and a low fraction undergoing mucociliary clearance. Apparent permeability in the intestine was predicted using the measured MDCK II permeability data multiplied by the scalar S _Papp determined using multiple reference compounds (cimetidine, atenolol, propranolol, verapamil, and metoprolol). Solubility in the GI tract is determined using measured aqueous solubility data together with the application of bile micelle‐mediated solubilization enhancement.

2.2.4. Systemic Clearance and Tissue Distribution

A combination of in vitro reaction phenotyping studies and human plasma metabolite identification studies showed the major metabolic pathways were direct glucuronidation (UGT2B7 and UGT1A9) and CYP‐mediated metabolism (CYP2D6, CYP2C9, and CYP3A4). The UGT and CYP clearances were then assigned and manually adjusted so that the ratio of the fraction metabolized by CYP enzymes to the fraction metabolized directly by UGTs matched that measured in the in vivo metabolite identification study with human plasma samples.

The volume of distribution was set by calculating the tissue partition coefficients as described by Rodgers and Rowland [21] multiplied by the K _p scalar parameter. Due to total clearance exceeding liver and kidney blood flow, additional systemic clearance was included in the model. Both K _p scalar and additional systemic clearance were set by optimization to the DPI clinical PK data.

2.3. Sensitivity Analysis

The sensitivity of plasma concentration to different lung regional deposition profiles was assessed by running six simulations for six different fold changes of upper and lower airway deposition for the 28 mg DPI dose ranging from ¼ to 3‐fold change. In each simulation, the total lung deposition is kept constant by setting the alveolar deposition as the total predicted lung deposition minus scaled airway deposition.

The sensitivity of the model to lung transport and dissolution parameters was assessed by global sensitivity analysis using the Morris method [28]. The analysis was performed for uniform distributions of the lung fluid solubility, dissolution scalar (diffusion layer model [DLM] scalar for lung dissolution rate with a default value of 1), fraction unbound in lung tissue, fraction unbound in lung fluid, and lung permeability.

The sensitivity of the PK profile to the CL parameterization (e.g., the UGT and CYP pathway and contribution assignment) was also assessed by setting all clearance to additional systemic clearance parameterized to achieve identical total IV clearance to the initial model. To maintain low oral bioavailability, the intestinal F _g was set to zero (represented in the model by setting ET deposition to zero).

3. Results

3.1. Lung Deposition

The predicted lung depositions for DP and Neb formulations are shown in Table 2. The total deposited dose for the DPI form is 100% of the emitted dose by model definition. A high proportion of the DPI dose is predicted to be deposited in the alveolar‐interstitial region (73%) with a total lung dose of 89% and 11% extrathoracic. The Neb dose is predicted to have 62% deposited in the extrathoracic region, 26% deposited in the lung regions (total of airway and alveolar), and 12% exhaled.

TABLE 2.

Predicted lung deposition percentages of delivered dose for both DPI and Neb dosage forms.

Region	%delivered dose deposited
	DP	Neb
ET	10.14	61.75
BB	13.57	5.19
bb	3.23	5.14
AI	73.06	15.54
Exhaled	0.0	12.37

Abbreviations: AI, alveolar‐interstitial; BB, bronchial; bb, bronchiolar; DP, dry powder; ET, extrathoracic; Neb, nebulized suspension.

3.2. Lung Absorption

The binding affinity to albumin (K _D) was back calculated from fu_plasma (0.0163) under the assumption that albumin is the only plasma protein that binds GDC‐6988 given that it is a neutral compound. The albumin K _D was estimated to be 11.29 μM. From this, fu_fluid was estimated to be 0.17. A small fraction of the dose deposited in the lung is lost due to mucociliary clearance. The maximum amount lost due to mucociliary clearance in the simulated individuals for the GDC‐6988 DPI dose was 2.6% of the total dose. For Neb dosing, less than 1% was lost to mucociliary clearance for all simulated individuals. The remainder of the deposited dose in the lung not lost to mucociliary clearance is absorbed. The mean lung fa for DPI dose and Neb dose was 0.89 and 0.25 for all doses, respectively. The difference in fa between DPI and Neb is due to the difference in the lung deposition as nearly all doses deposited in the lungs are eventually absorbed.

3.3. Intestinal Absorption

The apparent fractions of total dose absorbed via the intestines for the DPI and Neb doses are 0.10 and 0.61, respectively. This is greater than 99.9% of the dose deposited in the ET region that is assumed swallowed at the instant of administration. The predicted mean fraction of dose escaping gut metabolism (F _g) is approximately 0.005 for all doses and both dosage forms.

3.4. Optimized Systemic Clearance and Tissue Distribution Parameters

The refined model based on clinical PK data predicted a V _ss of 1.78 L/kg when a K _p scalar of 0.11 was applied. The optimized additional clearance was 194 L/h, resulting in an overall mean CL_iv of 235 L/h.

3.5. Plasma and Lung PK Simulations for DPI and Neb

Predicted and observed systemic concentration profiles for the DPI Phase 1b study (for model refinement) and Neb Phase 1 study (for model verification) are shown in Figures 2 and 3, respectively. Observed and predicted pharmacokinetic parameters (AUC, C _max, and T _max) are shown for the DPI Phase 1b study and the Neb Phase 1 study in Table 3. The absolute average fold errors (AAFE = ${10}^{\frac{1}{n}\sum \left|\log \left(\frac{{\mathit{\text{Pred}}}_i}{{\mathit{Obs}}_i}\right)\right|}$) for DPI predictions are 1.42 (AUC) and 1.27 (C _max) and for Neb predictions are 1.23 (AUC) and 1.45 (C _max). The predicted lung mass/concentration profiles in the upper airway and the alveolar‐interstitial regions for the DPI nominal dose of 28 mg are shown in Figure 4. GDC‐6988 is predicted to quickly dissolve in the lung fluid and permeate to the lung tissue, where it is retained for the whole dosing interval (i.e., about 12 h) and then decreases to a very low concentration at 12 h post dose in the upper airway. In the alveolar‐interstitial region, GDC‐6988 is retained for 1–2 h post dose.

FIGURE 2.

Observed and predicted plasma concentration profiles of GDC‐6988 for DPI clinical studies for nominal doses 11.2 mg (a), 28 mg (b), and 42 mg (c). The dots are observed values with each color representing an individual. The green solid lines are the mean model prediction. The gray lines are the 95th and 5th percentile confidence intervals of the simulated population. Each simulation used a healthy volunteer population of 100 individuals with a fraction of 0.5 females.

FIGURE 3.

Observed and predicted single‐dose plasma concentration profiles of GDC‐6988 for nebulizer clinical studies for nominal doses 22.5 mg (a), 45 mg (b), 75 mg (c), and 150 mg (d). The dots are observed values with each color representing an individual. The green solid lines are the mean population prediction for the refined model. The gray lines are the 95th and 5th percentile confidence intervals of the simulated population. Each simulation used a healthy volunteer population of 100 individuals with a fraction of 0.5 females.

TABLE 3.

Observed and predicted pharmacokinetic parameters for the DPI Phase 1b and Neb Phase 1 studies.

	Nominal dose	Statistic	Observed			Predicted			Fold error
			AUC0‐12 (h·ng/mL)	C max (ng/mL)	T max (h)	AUC0–12 (h·ng/mL)	C max (ng/mL)	T max (h)	AUC	C max
DPI	11.2 mg	Mean	34.4	30.1	0.167	20.31	39.81	0.07	0.59	1.32
		SD	11.9	11.3		4.21	13.27	0.02
	28 mg	Mean	62.1	70.9	0.083	50.60	98.50	0.07	0.81	1.39
		SD	16.9	15.8		10.49	32.82	0.03
	42 mg	Mean	105.40	130.47	0.083	75.66	147.06	0.08	0.72	1.13
		SD	18.46	49.59		15.67	49.15	0.03
Neb	22.5 mg	Mean	8.46	20.30	0.01	8.08	11.96	0.05	0.96	0.59
		SD	12.18	12.43		1.82	3.49	0.02
	45 mg	Mean	18.11	46.05	0.01	16.97	25.109	0.05	0.94	0.55
		SD	12.18	32.77		3.81	7.34	0.02
	75 mg	Mean	20.89	43.3	0.17	32.06	44.64	0.04	1.54	1.03
		SD	11.63	24.16		7.37	11.21	0.02
	150 mg	Mean	80.02	101.53	0.46	59.00	71.81	0.04	0.74	0.71
		SD	9.64	42.01		14.01	15.49	0.01

FIGURE 4.

Predicted lung mass/concentration profiles for the DPI nominal dose of 28 mg. The plots are the amount of undissolved mass (a, d); the concentration in the fluid compartment (b, e); the concentration in the tissue mass compartment (c, f). Panels (a, b, and c) represent the upper airway region and panels (d, e, and f) represent the alveolar‐interstitial region. The green solid lines are the mean population prediction. The gray lines are the 95th and 5th percentile confidence intervals of the simulated population. Each simulation used a healthy volunteer population of 100 individuals with a fraction of 0.5 females.

3.6. Sensitivity Analyses

The predicted profiles for changes in the ratio of airway (BB + bb) to alveolar (AI) deposition are shown in Figure S1. Decreasing/increasing airway deposition leads to an increase/decrease in C _max. An airway deposition of 0.25‐fold of base predictions leads to a 16% increase in C _max, and a 3‐fold increase in airway deposition leads to a 45.2% decrease in C _max. The T _max reported for all simulations is identical. The AUC_0–12 changes by < 2.5% compared to base model predictions for all simulations.

Global sensitivity analysis for dissolution and lung transport parameters showed that AUC and C _max are sensitive to the parameters of lung permeability, fu_mass, fu_fluid, lung solubility, and dissolution scalar included in the analysis (Figure S2). The impact of parameters is ranked by their mean elementary effect (μ*). The AUC and C _max were most sensitive to changes in the fraction unbound in lung tissue, followed by the basal permeability, with μ* for C _max 20‐fold higher than for AUC. The fraction unbound in lung tissue and lung fluid solubility were similarly ranked, with μ* roughly half of the top ranked parameters. AUC and C _max were less sensitive to changes in dissolution scalar and apical permeability, with a μ* roughly a tenth of the top ranked parameter.

The impact of assigning part of the clearance to UGT and CYP was compared to assigning all clearance as additional clearance, and it showed relatively little impact on the systemic PK profiles (Figure S3).

4. Discussion

This manuscript describes a mechanistic lung PBPK model with key components of lung deposition, lung and intestine absorption, systemic clearance, and tissue distribution included. This work provided an opportunity to establish a PBPK modeling framework that has the potential to predict the plasma and regional lung PK to perform meaningful exposure–response analysis and inform dose selection in the clinical development of GDC‐6988 and future inhaled investigational drugs.

The PBPK model of GDC‐6988 was parameterized using a combination of a bottom‐up and top‐down approach. The majority of initial input parameters were set by in vitro to in vivo extrapolation (IVIVE). The distribution and clearance were optimized against the DPI clinical plasma PK data. The model was then used to predict the Neb study PK without any further optimization. This model's predictions gave reasonable agreement with the Neb observations, although it underpredicted individual variability. The current deposition model produces a prediction for males and females based on sex differences in physiological parameters, but there is no variability between individuals of the same sex. This is a limitation of the current model, as variability in lung physiology has been observed to be dependent on other covariates such as individual age and size. Incorporation of these covariate relationships in the future may make the simulated level of variability closer to the observed data [29, 30]. There is much less variability with the DPI clinical observations, and the variability is similar to that from the model predictions. Following DPI administration, the majority of the emitted dose is predicted to be deposited in the alveolar region of the lung with subsequent rapid absorption, resulting in lower individual differences in deposition in the clinical data.

For the DPI study, the model gives reasonable predictions for early time points (< 30 min) and late time points (≥ 4 h), but there is an underprediction in concentration between 40 min and 1 h time points resulting in an underprediction of AUC (of 41%, 19% and 28% for the three doses, Figure 2, Table 3). This is not seen with the Neb study (Figure 3, Table 3). It is possible that DP dissolution is not being predicted accurately. Another explanation is that regional deposition is better predicted for Neb. The results of an updated model where parameters for lung deposition, dissolution, and transport were further optimized are shown in Section S.4: Data S1 and Table S7, Figures S5 and S6. This model has higher deposition in the airways and a 10‐fold slower dissolution rate. There is a closer agreement to all of the time points with the alveolar deposition driving the C _max achieved in the first 10 min and the slower dissolution in the airways maintaining the concentration (and avoiding the underprediction of AUC) between the 40 min to 1 h observations. This suggests a two‐phase lung absorption that is not captured by the original model. One assumption in the dissolution model is that particles are immersed in lung fluid at the moment of deposition, which may lead to overprediction of the dissolution rate. Adapting this assumption to include partial particle fluid immersion and particle wetting could be investigated in the future. The model then was verified with the Neb data and still achieved reasonably good AAFE for AUC (1.23) and C _max (1.46), which suggests the model is capturing transport within the lung and systemic distribution and clearance.

The predicted AUC is typically not sensitive to changes in lung parameters. This is expected as the majority of the dose deposited in the lung is absorbed for parameter changes (e.g., dissolution rate or permeability) up to 10‐fold. For GDC‐6988, mucociliary clearance is low (< 3% of dose), and no other elimination pathways (e.g., metabolic clearance) are included in this model.

The C _max and shape of profile are sensitive to changes in regional deposition (Figure S1) and lung absorption parameters (Figure S2). The C_max is most sensitive to tissue fraction unbound and permeability. The rate of absorption also impacts the C _max and T _max. The rate of dissolution is different in the airway and alveolar regions due to the difference in fluid volume and the low solubility of GDC‐6988. The relative differences in surface area and blood flow between the airway and alveolar regions also impact the absorption rate. This means determining the regional deposition can be important for estimating compound PK.

While the mechanistic model can help overcome the challenge of sampling human lung tissue to estimate lung PK, numerous assumptions have to be made in the PBPK model, which leads to uncertainty for the plasma and lung PK prediction. To reduce the uncertainty, robust estimates of key parameters in the model, like oral absorption, systemic clearance, and volume of distribution, can be of great help. Confidence in model predictions could be increased if a human ADME study with IV and oral routes of administration can be conducted to obtain those parameters directly without dependence on lung deposition and absorption [31].

Compared to other absorption routes (e.g., oral and dermal), verifying models of lung absorption has more challenges. Accurate measurements of regional lung deposition are lacking. In vitro models for predicting permeability and absorption continue to evolve [32] but are currently less well established. In this study, permeability was estimated using a QSAR model which takes compound log D as an input. Given the sensitivity of the C _max to permeability, as shown by the local sensitivity analysis (Figure S4) it would have been preferable to use an in vitro measure of permeability. A correlation to estimate lung permeability from Caco‐2 permeability is available [13] but a more robust permeability measurement would use lung‐specific cells such as calu‐3. There is also uncertainty about the correlation between impactor measurements and in vivo deposition [33, 34]. To increase confidence in model predictions, the lack of data and established in vitro methods will need to be addressed.

One thing that could benefit the future application of the model (e.g., dose selection) is to include individual variability in physiology parameters that impact deposition in this lung PBPK model. Ideally, this would be assessed against clinical measurements of regional deposition for individuals with measured lung function. Assessing the impact of more physically representative immersion of particles in fluid on predicted dissolution could also be done to determine whether this is the cause of the potential underprediction of intermediate observations.

This mechanistic lung PBPK model also has the potential to be used to predict plasma and lung PK for formulation/device change (e.g., Neb to DPI, DPI formulations of different particle size properties) with lung deposition adjustments based on device and particle size properties. It may also be used to predict PK and support dose selection for population change (e.g., healthy volunteers to patients) with lung deposition differences predicted via functional respiratory imaging (FRI) approach (i.e., combined high‐resolution computed tomography (HRCT) scans with Computational Fluid Dynamic technique) [7] and also with changes in physiological parameters that affect lung absorption (e.g., protein binding in airway fluid) between the two populations. Besides respiratory diseases, this PBPK modeling framework can also benefit the development of molecules in other indications, like oncology, via predicting tissue distribution to facilitate clinical indication selection and assessing systemic exposure for both pharmacology and toxicology purposes [5].

Author Contributions

I.S., R.Z., Y.C., F.M., M.N., M.J., and I.G. wrote the manuscript. R.Z., Y.C., F.M., M.N., N.K., and R.O. designed the research. R.Z., F.M., Y.C., J.G., T.D.B., and Y.‐J.S. performed the research. I.S., R.Z., F.M., and G.S. analyzed the data.

Conflicts of Interest

R.Z., F.M., M.N., Y.‐J.S., G.S., T.D.B., J.G., N.K., R.O., and Y.C. are employees of Genentech Inc., a member of the Roche group, and Roche shareholders. I.S., I.G., and M.J. are employees of Certara UK Limited.

Supporting information

Data S1.

Data S2.