Computational Fluid Dynamics (CFD) Guided Spray Drying Recommendations for Improved Aerosol Performance of a Small-Particle Antibiotic Formulation

Abstract

Purpose:

The objective of this study was to implement computational fluid dynamics (CFD) simulations and aerosol characterization experiments to determine best-case spray drying conditions of a tobramycin excipient enhanced growth (Tobi-EEG) formulation for use in a pediatric air-jet dry powder inhaler (DPI).

Methods:

An iterative approach was implemented in which sets of spray drying conditions were explored using CFD simulations followed by lead candidate selection, powder production and in vitro aerosol testing. CFD simulations of a small-particle spray dryer were performed to capture droplet drying parameters and surface-averaged temperature and relative humidity (RH) conditions in the powder collection region. In vitro aerosol testing was performed for the selected powders using the pediatric air-jet DPI, cascade impaction, and aerosol transport through a pediatric mouth-throat (MT) model to a tracheal filter.

Results:

Based on comparisons of CFD simulations and in vitro powder performance, recommended drying conditions for small-particle powders with electrostatic collection include: (i) reducing the CFD-predicted drying parameters of κ_avg and κ_max to values below 3 µm²/ms and 114 µm²/ms, respectively; (ii) maintaining the Collector Surface RH within an elevated range, which for the Tobi-EEG formulation with l-leucine was 20–30 % RH; and (iii) ensuring that particles reaching the collector were fully dried, based on a mass fraction of solute CFD parameter.

Conclusions:

Based on the newly recommended spray dryer conditions for small particle aerosols, delivery performance of the lead Tobi-EEG formulation was improved resulting in >60% of the DPI loaded dose passing through the pediatric MT model.

INTRODUCTION

Spray drying provides a flexible, controllable, and highly scalable method to produce pharmaceutical aerosol powder formulations (1–6). When combined with the principles of particle engineering, spray-dried formulations can be designed to enhance aerosolization performance, stability, bioavailability and efficacy (3, 4, 7–13). While highly flexible, challenges with spray drying include the large number of input variables and complex interdependent relationships that impact formulation properties. For example, elevated drying temperature can significantly enhance powder dispersibility in some instances and decrease dispersion performance in others, likely dependent on initial droplet size, droplet constituents, shell formation and crystallization (1, 10, 14, 15). To address these challenge, multiple forms of modeling are currently used to better understand and design effective spray drying processes that produce pharmaceutical powders with specific attributes (1, 2, 8, 14).

Systems of interconnected ordinary differential equations (ODEs) are frequently implemented as a modeling strategy to predict droplet drying kinetics, radial distribution of solutes in the dried particles and final surface concentrations of constituents (1, 2, 8, 15–18). The use of ODE systems to understand droplet drying was originally developed for combustion sprays (19) and has been applied to spray drying of food products for decades (20, 21). This approach was adopted and advanced for the spray drying of pharmaceutical powders notably by Vehring et al. (1, 2), and has been applied in a series of excellent studies on pharmaceutical aerosols (8, 15–18). Using this approach, drying and droplet formation parameters can be predicted that enable radial control of droplet constituents and surface enrichment of drugs or dispersion enhancers. Identified non-dimensional numbers can be used to predict the type of particle structure that will likely form and serve to explain experimental findings. Based on these expressions, higher evaporation rates are often preferred to create highly dispersible powders; however, sufficient crystallization time is also needed in the presence of potential crystalline dispersion enhancers (8, 15–18). The ODE system approach has recently been advanced to better understand and predict the spray drying behavior of nebulized formulations with multiple solvents (e.g., water and ethanol) (22). While effective in many respects, a limitation of the ODE approach is the general assumption of a constant drying rate across a droplet’s lifetime and the assumption that all droplets in a spray dryer experience the same drying conditions.

Computational fluid dynamics (CFD) is a numerical simulation technique that fully resolves transport phenomena in three-dimensional (3-D) space and across time, based on discretization of the flow field and numerical solution of interconnected systems of partial differential equations (PDEs) (23). Simulation of the flow field can be interconnected with discrete droplet elements that are modeled to evaporate and solidify, typically implementing ODE submodels (19, 24, 25) that are coupled to the continuous field (26–28). Previous review articles have highlighted the use of CFD to study the general drying kinetics with industrial spray dryers (29). In these previous CFD studies (29, 30), numerically predicted particle sizes showed relatively poor agreement with experimentally measured primary particle diameters, and predictions of aerosol size from a DPI were not attempted. The study of Longest et al. (14) was the first attempt to link CFD predictions of spray drying with aerosolization performance of the resulting powders. The CFD model captured the spray drying conditions of the previous experimental study of Son et al. (10) with the evaluation of an albuterol sulfate (AS) excipient enhanced growth (EEG) formulation containing l-leucine as a dispersion enhancer. Simulations captured both the two-way coupling between the gas and droplet phases and the spray momentum introduced from acoustic streaming arising from the mesh nebulizer. Drying parameters included a time-averaged drying rate (κ_avg), maximum instantaneous drying rate (κ_max) and solute precipitation window (t_p), each averaged across all droplets in the system. High variability in drying rate was observed across droplets in each system, and instantaneous drying rate was often two orders of magnitude higher than the average drying rate for individual droplets. Surprisingly, reducing κ_avg and κ_max improved aerosolization behavior when the powders were tested in a reference capsule-based dry powder inhaler (DPI) with a reduction in mass median aerodynamic diameter (MMAD) and an increase in device emitted dose (ED). It was postulated that κ_max values that were excessively high prevented l-leucine shell formation and potentially reduced crystallization (14).

The recent experimental spray drying study of Hassan et al. (31) sought to improve the stability and aerosolization performance characteristics of an EEG formulation of tobramycin that also included l-leucine as a dispersion enhancer. Drying was performed using inlet air with water vapor content levels of 1 or 10 g/m³. Surprisingly, the higher water vapor content level of 10 g/m³ (equal to RH of ~50% at standard room conditions) produced powders with higher crystallization and glass transition relative humidities (%RH) indicating increased stability. These results provide further evidence that with small-droplet spray drying systems, like the Buchi Nano B-90 HP Spray Dryer, reducing the drying rate can improve formulation performance.

Considering the previous studies of Longest et al. (14) and Hassan et al. (31) together raises a number of questions that may lead to improved formulation optimization. For example, do different input parameters to reduce droplet drying rate (e.g., reduced drying temperature or increased drying air RH) have the same effect on powder aerosolization performance? Can the effects of reduced drying temperature and increased drying gas water vapor content be captured in a single drying parameter and correlated with experimental data? At what point does increased drying gas water vapor content effect powder performance and can this effect be captured in CFD simulations and associated performance correlations? Finally, what are CFD-predicted optimization conditions for producing tobramycin–EEG formulations with the Buchi Nano B-90 HP Spray Dryer and do these conditions further improve aerosol dispersion and lung delivery efficiency when administered with a new pediatric air-jet DPI?

The objective of this study is to implement CFD simulations and aerosol characterization experiments to determine best-case spray drying conditions of a specific Tobi-EEG formulation for use in a new pediatric air-jet DPI. The analysis is conducted across multiple stages with a first stage considering the best-case spray drying conditions and Tobi-EEG formulation determined experimentally by Hassan et al. (31) compared with CFD-based rapid screening of eight new drying conditions. Using these CFD results, four sets of spray drying conditions are selected for implementation, aerosol characterization and development of correlations between CFD predicted drying parameters and experimental results. Additional simulations are then conducted that use the new drying correlations and other CFD insights to guide the development of a lead set of spray drying conditions. Final experiments are then conducted to evaluate both the aerosolization performance and in vitro lung delivery efficiency of the lead case. Results of this study are intended to demonstrate the use of CFD modeling to assimilate a broad and complex set of spray drying input parameters for the prediction of aerosolization performance and to improve the lung delivery efficiency of an antibiotic spray-dried formulation intended for administration to pediatric and adult subjects.

MATERIALS AND METHODS

Spray Drying Conditions

The Buchi Nano B-90 HP Spray Dryer (Buchi Laboratory-Techniques, Flawil, Switzerland) was evaluated in both the CFD and experimental components of this study, as represented in Figure 1a. Unique features of the B-90 HP spray dryer include uniform inlet laminar drying gas flow, a vibrating mesh nebulizer for droplet production, a cylindrical drying chamber with a length and diameter of 75.3 and 16.4 cm, respectively, and an electrostatic precipitator for efficient collection of small particle aerosols on the outer wall of the collection region (32). As with Hassan et al. (31), an excipient enhanced growth formulation of the antibiotic tobramycin was nebulized in a 80:20%^v/_v vehicle of water and ethanol using the medium droplet size vibrating mesh in all cases. The nebulized formulation was consistent with Case 5 of Hassan et al. (31) with tobramycin, l-leucine, mannitol, and poloxamer 188 included in a mass ratio of 60:20:18:2, respectively, with a total solids concentration in the spray solution of 0.3% w/v. This formulation was implemented in all cases of the current study. Base spray drying conditions were also consistent with Case 5 of Hassan et al. (31) including a drying gas flow rate of 120 LPM, 70 °C inlet drying temperature and drying gas water vapor content of 10 g/m³ pre-nebulization. CFD simulations began with Case 5 conditions from Hassan et al. (31) and then varied inlet drying temperature (35 to 70 °C), drying gas flow rate (80 to 160 LPM) and drying gas water vapor content (1 to 10 g/m³). At a standard laboratory temperature of 24 °C, drying gas water vapor contents of 1 and 10 g/m³ equate to RH values of 4.6 and 46.4%, respectively. Based on previous experiments with the B-90 medium mesh and similar ionic solids concentrations (14), the nebulized mass median droplet size was assumed to be 6.13 µm with a nebulization rate of 0.6 ml/min, which were held constant and implemented in all CFD simulations.

Figure 1.

CFD geometry of the Buchi Nano B-90 HP Spray Dryer including (a) 3D rendering with dimensions and (b) midplane CFD-predicted velocity magnitude and vectors illustrating the acoustic streaming effect from the mesh nebulizer.

CFD Simulations: Computational Geometry, Mesh and Boundary Conditions

A computational geometry of the Buchi B-90 HP spray dryer was constructed as previously described by Longest et al. (14). The geometry was discretized with hexahedral elements with significant grid refinement near the walls and mesh nebulizer. While the previous computational geometry ended at the start of the electrostatic precipitator, this region was included in the current study in order to predict temperature and RH exposure conditions on the surface of the collector walls during spraying. This region extended the computational geometry by 32 cm and contained an internal charging electrode geometry with a diameter of 9.5 cm (Figure 1a). This inner electrode was approximated with a solid cylinder to avoid the complexity of the multiple serrated disks implemented in the physical model. Moreover, the electrostatic field of the collection region was not simulated as it is not anticipated to influence the temperature and RH of the collection region, which were a focus of this study. Finally, flow was allowed to exit the bottom ring formed by the outer wall and electrostatic precipitator inner cylinder (Figure 1a).

The computational mesh contained 1,836,960 hexahedral control volumes with wall y+ values of approximately 1. The region of the vibrating mesh in the mesh nebulizer spray head (with a physical spray diameter of 8 mm) was meshed with 32 elements across the diameter and 120 elements around the perimeter. Increasing the mesh cell count by 50% throughout the spray dryer had a negligible effect (<5% relative difference) on all drying parameters considered. At the top inlet of the spray dryer, nearly uniform flow (with a smooth transition to zero wall velocity) was implemented. Based on an inlet volumetric flow rate of 120 LPM (2000 cm³/sec), the inlet Reynolds number was approximately 870, indicating laminar conditions. No slip was assumed on all surfaces and particles were assumed to deposit upon initial surface contact. A constant pressure boundary condition was assumed across the lower outlet as displayed in Figure 1a.

To better approximate temperature and RH conditions in the collection region of the spray dryer, heat loss through the wall boundaries was taken into account in this study. The spray dryer walls of the Buchi B90 HP are composed of borosilicate glass with a thickness of approximately 8.7 mm. The CFD thermal boundary condition was a local heat flux (W/m²) defined as

$$q^{″}=\frac{\left(T_{surf,inner}-T_{\infty }\right)}{\frac{x_{wall}}{k_{wall}}+\frac{1}{h_{\infty }}}$$ (1)

where x_wall is the spray dryer wall thickness, k_wall is the wall thermal conductivity coefficient (1.4 Wm⁻¹K⁻¹, $h_{\infty }$ is the convective heat transfer coefficient between the outer wall boundary and the environment, T_surf,inner is the temperature of the inner wall surface, and $T_{\infty }$ is the temperature of the surrounding environment. The convective heat transfer coefficient describing conditions between the outer spray dryer wall and surrounding environment $\left(h_{\infty }\right)$ was taken from the Nusselt number expression for a vertical cylinder experiencing natural convection resulting in an average value of approximately 5 Wm⁻²K⁻¹ (33). The entire length of the spray dryer (1.18 m) outer wall was assumed to be experiencing natural convection based on its higher temperature compared to the surrounding environment. Use of an environmental temperature of $T_{\infty }=21{\hspace{0.17em}}^{\circ }\text{C}$ provided good agreement between the CFD-predicted and measured spray dryer outlet temperatures (within 10% relative difference).

As with the previous study of Longest et al. (14), acoustic streaming from the mesh nebulizer into the flow field was simulated using the expression of Trujillo and Knoerzer (34), which introduces an air jet into the velocity field with a radial (r) velocity defined as

$$V\left(r\right)=\left(\frac{2K}{{\rho }^2\pi S^2}\right)e^{\left(-r/S\right)}$$ (2)

where K represents the kinematic momentum of the jet, ρ is the fluid density, S is a jet width scale factor, and r is the radial coordinate. It was assumed that the maximum acoustic streaming jet velocity was linearly proportional to the nebulized liquid mass flow rate, ${\dot{m}}_l$. As a result, the previous conditions for ${\dot{m}}_l=0.4\hspace{0.33em}\text{ml}/\min$ from Longest et al. (14) were adjusted to match a measured ${\dot{m}}_l=0.6\hspace{0.17em}\text{ml}/\min$ flow rate in this study producing a peak inlet velocity of 12 m/s. The resulting parameters for the acoustic streaming velocity equation were S = 5E-3 m, K = 7.8E-5 kg²m⁻²s⁻², and r = 4E-3 m. Because the diameter of the vibrating mesh that produces aerosol is only 8 mm, the injected additional mass to capture the streaming effect was negligible, accounting for less than 2% of the total mass flow rate in the system. Turbulence properties at the vibrating mesh inlet were assumed to be 5% intensity and 5% turbulent viscosity ratio.

Velocity conditions within the B-90 HP spray dryer with a drying air flow rate of 120 LPM and acoustic streaming arising from the mesh nebulizer are illustrated in Figure 1b. As observed in the figure, the presence of the mesh nebulizer and acoustic streaming result in high velocity jet flow and a turbulent flow environment. The presence of this turbulent jet will result in turbulent mixing within the spray dryer and turbulent dispersion of the droplets, as illustrated in Longest et al. (14).

CFD Simulations: Transport Equations and Numerical Simulations

In order to capture the turbulent and transitional flow field within the spray dryer, the two-equation Low Reynolds Number (LRN) k-ω turbulence model was implemented, which is capable of transitioning between laminar and turbulent flow in a single geometry (25, 35). Use of this model in transitional and turbulent flow fields has previously demonstrated good agreement with experimental data describing pharmaceutical aerosol size change (27, 36) and deposition (25, 36–39). The governing transport equations for conservation of mass and momentum, turbulent kinetic energy (k) and specific dissipation rate (ω) are available in our previous publications (25, 36).

The continuous phase gas was composed of air, water vapor (w) and ethanol vapor (eth) components. In all equations, density of this gas mixture was determined using the multicomponent ideal gas law, which can be written as

$$\rho =\frac{P}{R_{u}T\left(\frac{Y_{air}}{M_{air}}+\frac{Y_w}{M_w}+\frac{Y_{eth}}{M_{eth}}\right)} .$$ (3)

In this equation, P is the total gas pressure, R_u represents the universal gas constant, Y_s is the mass fraction of each gaseous species (air, water vapor and ethanol), and M_s is the molecular weight of each species.

Governing equations describing the mass transport of each vapor component and the transport of thermal energy for a multiple component gas in turbulent flow were previously presented in Longest et al. (14). These equations implemented source/sink terms to capture the transfer of energy between the droplet and continuous phases as the droplets evaporated (i.e., two-way coupling). This process reduces the temperature of the continuous phase due to the latent heat of vaporization of the droplets and increases the mass fraction of water and ethanol in the gas mixture. The resulting RH values in the continuous phase were calculated as

$$RH=\frac{P_w}{P_{w,sat}}=\frac{Y_w \rho R_{w}T}{Y_{w,sat}\rho R_w T}=\frac{Y_w}{Y_{w,sat}}.$$ (4a)

$$R{H}_{eth}=\frac{P_{eth}}{P_{eth,sat}}=\frac{Y_{eth}\rho R_{eth} T}{Y_{eth,sat} \rho R_{eth} T}=\frac{Y_{eth}}{Y_{eth,sat}}.$$ (4b)

In these expressions, P_s, Y_s and R_s represent the partial pressure, mass fraction and gas constant of each species (s). The subscript sat is used to denote Saturation conditions. Saturation pressures for water and ethanol vapors (P_w,sat and P_eth,sat) were calculated using the Antoine equation with coefficients from Green (40), as described in our previous studies (36).

Lagrangian tracking was used to simulate droplet and particle motion included terms for discrete element drag, gravity, and buoyancy (25). A random-walk eddy interaction model (41) was used to simulate turbulent dispersion, with equations presented in Longest et al. (25). As with our previous studies, near-wall corrections including turbulent anisotropic effects and particle-wall hydrodynamic interactions were applied at all wall boundaries (25, 42), and droplets were assumed to deposit upon initial wall contact. Droplet agglomeration was not included.

A numerically efficient rapid mixing model (RMM) was implemented to capture droplet drying, which assumes a uniform temperature profile and uniform mass concentrations within the droplet (43). Implementation of this approach for the small droplet and spray dryer conditions considered was based, in part, on the study of Ordoubadi et al. (22), which demonstrated little difference between the RMM assumption and more detailed ODE drying models at moderate drying temperatures as considered in this study. Furthermore, Feng et al. (18) predicted relatively low levels of l-leucine surface enrichment when using the B-90 spray dryer. While the RMM is expected to work well for the droplet solution formulation and drying conditions considered in this study, this assumption may not be valid for suspension formulations or with significantly elevated evaporation rates.

Governing equations for the droplet conservation of mass and energy with the rapid mixing model were reviewed in Longest et al. (14). Expressions from Clift et al. (44) were used for the Nusselt and Sherwood numbers to describe droplet heat and mass transfer, respectively. In calculating droplet surface properties, saturation pressures were again determined from the Antoine equation with coefficients from Green (40) and gas mixture density was determined from the ideal gas law, Eq. (4). Water activity coefficients were based on a multicomponent version of Raoult’s law (14, 26). Due to the relatively low solubility of all components in ethanol, all solutes were assumed to only be dissolved in the aqueous component of the initial 80:20 %^v/_v water:ethanol solution. Future studies may need to consider the impact of solubility in binary water:ethanol solvents (45).

Two-way heat and mass coupling between the phases includes the thermodynamic effects of the droplets on the gas field. As droplets evaporate, two-way coupling results in gas phase cooling as well as an increase in gas phase water and ethanol vapor content, which reduce the droplet drying rate. Based on the previous CFD study of Longest et al. (14), two-way coupling is known to be an important factor during spray drying in the Buchi B-90 system. As described previously (14, 27), two-way coupling was included in the simulations using a droplet parcel approach. Briefly, 1500 droplet parcels were used to represent the computational discrete field. Each parcel represented a collection of droplets, and for initially monodisperse droplets under steady state conditions, each parcel contained 5.54E4 droplets to capture the injected liquid mass from the mesh nebulizer. The CFD solution gradually included the influence of the discrete phase on the continuous phase through a series of iterations between solutions of the respective fields. A total of 40 iterations with an under-relaxation factor of 0.05 was found to fully capture the influence of two-way heat and mass transfer coupling.

CFD Simulations: Drying Parameters

Based on the previous CFD analysis of small particle spray drying performed by Longest et al. (14), drying parameters that were predictive of aerosolization performance included the average droplet evaporation rate (κ_avg), maximum instantaneous evaporation rate (κ_max), time integral of the saturation ratio (ISR), and precipitation window (t_p). At each time point during droplet evaporation, an instantaneous evaporation rate can be calculated as

$${\kappa }_{inst}\left(t\right)=\frac{d}{dt}\left(d_{droplet}^{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}2}\right)\cong \frac{d_{prev}^2-d_{new}^2}{dt}$$ (5)

with units of µm²/ms where d_droplet is the current geometric diameter of the drying droplet, dt represents a small segment of sampling time (i.e., time step), and d_prev and d_new are the droplet geometric diameters over the time step. A time average of the instantaneous evaporation rate can be calculated as

$${\kappa }_{avg}=\frac{1}{t_{dry}}{\displaystyle {\int }_0^{t_{dry}}{\kappa }_{inst}\left(t\right)dt}$$ (6)

where the time integral is evaluated between initial nebulizer release of the droplet (t = 0) and the point at which the total droplet solute mass fraction $\left(\sum Y_s\right)$ reaches ≥0.98 (defined in this study as t_dry). Each droplet trajectory results in one κ_avg value. For each spray drying case, the median of these κ_avg values was calculated and taken as the representative κ_avg value of that case.

As with the average evaporation rate, a maximum instantaneous evaporation rate for each droplet trajectory can be calculated as

$${\kappa }_{max}=max\left[\frac{d_{prev}^2-d_{new}^2}{dt}\right].$$ (7)

For each spray drying case, the median of these values was calculated and taken as the representative κ_max value.

Similar to a droplet crystallization window (1, 2), a droplet precipitation window (t_p) can be defined as the time available for drying to occur at high solute concentrations (above the saturation condition):

$$t_{p,i}=t_{dry}-t_{sat,i}.$$ (8)

As specified above, t_dry represents the time from initial release to the point when the droplet is ≥98% dry by mass. Saturation time (t_sat,i) is the time from droplet release until a specific solute (here l-leucine) reaches saturation. The precipitation window therefore represents the time period between initial saturation of l-leucine in the droplet and the when the droplet is almost completely dry.

Finally, a time integration of the saturation ratio can be calculated as

$$ISR={\displaystyle {\int }_{t_{sat,i}}^{t_{dry}}\frac{m{f}_i\left(t\right)}{m{f}_{i,sat}}dt}$$ (9)

where mf_i represents the droplet mass fraction of solute i at time t. Dividing this value by the saturation mass fraction of the same solute (mf_i,sat) forms a saturation ratio. Integration of the saturation ratio over time provides an indication of the time a droplet spends in a saturated state together with the level of supersaturation. As with other dispersion parameters, ISR values were computed for each droplet in a specific spray drying case and the median value was taken as the representative ISR value for that case.

While spray drying often produces hollow, porous and/or wrinkled particles, calculation of a theoretical solid geometric particle diameter can be useful, defined as

$$d_{final}=d_{initial}{\left(\frac{{\overline{\rho }}_{initial}}{{\rho }_{final}}\right)}^{1/3}.$$ (10)

In this expression, ${\overline{\rho }}_{initial}$ and ρ_final represent the initial bulk density (or solute concentration expressed as weight/volume) and final solid particle density, respectively. Based on solute concentrations in the initial droplets for the Tobi-EEG formulation, the corresponding final density of the theoretical solid particles (ρ_final) is 1.46 g/cm³.

Experimental: Production and Storage of Spray-Dried Powders

For selected cases, the associated spray drying conditions were implemented and the powders were produced using the Buchi Nano B-90 HP Spray Dryer. Condition ranges and the spray vehicle formulation content were described in the Spray Drying Conditions section. Spray drying methods were previously described in Hassan et al. (31). Briefly, water vapor content of the drying gas was controlled by humidifying filtered room air in a pre-conditioning chamber prior to the heated B-90 inlet. Nebulized liquid flow rates were in the range of 0.57 to 0.73 ml/min and could not be better controlled. Approximate spray times were 180 min and average powder batch sizes were approximately 300 mg. After spray drying was completed, the powders were recovered from the collection region, placed in brown glass vials and covered with a gas permeable material. The covered glass vials were stored under controlled temperature/RH and given a one week relaxation period before experimental testing. Samples were stored at 21±2°C and 30±3%RH. It is not expected that effects of this RH exposure penetrated beyond the top layer of powder in the covered glass vials.

Experimental: Pediatric DPI

Aerosolization performance of the spray-dried powders was tested with a new pediatric air-jet DPI as a reference device. As described in Farkas et al. (46), the pediatric air-jet device includes a small diameter air inlet, vertical aerosolization chamber and small diameter flow passage outlet (Figure 2). The turbulent aerosol jet flow exiting the central aerosolization unit is dissipated with a 3D rod array structure in the mouthpiece (46–48). The design selected for this study was Device-3 coupled with Mouthpiece-2, which included inlet and outlet flow passage diameters of 1.4 and 2.39 mm, respectively (46). The pediatric air-jet design is operated with a custom positive-pressure gas source that generates the aerosol and provides a full inhalation breath to the subject. The positive-pressure gas source was set to provide a device inlet pressure of 6 kPa with a square waveform and an inhaled volume of 750 ml, which is 75% of vital capacity for a 5-year-old child (49), at a flow rate of approximately 10 LPM. Powder is loaded into the removable lower portion of the pediatric air-jet device, which is then reconnected to the DPI using a quarter-turn twist lock mechanism.

Figure 2.

Rendered 3D drawing of the pediatric air-jet DPI used for aerosolization testing. Key components include the small diameter inlet flow passage, vertical capsule-shaped aerosolization chamber, outlet capillary and 3D rod array used to dissipate the air-jet and further deaggregate the powder.

Experimental: Lung Delivery Estimate using the Pediatric Mouth-Throat (MT) Model

Lung delivery of the spray-dried formulations was estimated using a pediatric MT model of a representative 5-year-old subject (Figure 3). As described in the study of Farkas et al. (46), the pediatric MT model implemented the VCU Medium Adult geometry (www.rddonline.com) and scaled this geometry down using a linear factor of 0.75. This scale factor was selected in order to match the tracheal outlet diameter to the mean 5-year-old value reported by Phalen et al. (50). For connection with the pediatric air-jet DPI, the MT opening was given the shape of a 1.7 by 2.2 cm ellipse, which was then smoothly blended to the rest of the MT geometry. As described in more detail in Farkas et al. (46), the resulting MT geometry had internal surface area and volume values of 63.9 cm² and 26.2 cm³, respectively, with key dimensions in-line with anatomical studies and other pediatric MT models.

Figure 3.

Pediatric MT model used for lung delivery testing with downstream tracheal filter to capture the lung delivered dose.

Lung delivery efficiencies of the initial Case 5 powder and the new best-case powder were estimated using the pediatric MT model using the pediatric air-jet DPI loaded with 10 mg fill masses of the formulations. The MT model used in this study was built with 3D printing using Accura ClearVue resin. As shown in Figure 4, the air-jet DPI and mouthpiece (MP) were connected to the MT geometry. To minimize particle bounce and re-entrainment, the MT was coated with MOLYKOTE^® 316 silicone spray (Dow Corning, Midland, MI). In order to better capture the small particle aerosol formed with the spray-dried formulation and pediatric air-jet DPI, two low resistance filters (Pulmoguard II) were connected in series to the tracheal outlet of the MT model, and aerosol deposition on these filters was used to approximate lung delivery efficiency.

Figure 4.

Profile view of the pediatric air-jet DPI with mouthpiece (MP) connected to the pediatric mouth-throat (MT) geometry.

Drug deposited on the MT model and filters was collected following washing with deionized water. Tobramycin concentrations in the solutions were determined by a validated liquid chromatography-mass spectrometry (LC-MS) method (31). Drug recovery was expressed as % of the loaded dose. Overall drug recovery was determined as the sum of deposition on all components (air-jet DPI, MP, model, and filter) expressed as a % of the loaded tobramycin dose.

Experimental: Evaluation of Formulation Aerosol Characteristics from the Air-Jet DPI

To determine the aerosolization performance of selected spray-dried powders, the pediatric air-jet DPI (loaded with 10 mg fill masses of Tobi-EEG powder) was connected to the preseparator inlet of a Next Generation Impactor (NGI). The NGI was oriented vertically to allow the air-jet DPI to be held horizontally, as it would be in practice. As illustrated in Figure 5, a custom adaptor was used to allow co-flow air from the environment to surround the device emitted aerosol. To provide adequate discrimination of the small-particle aerosol, the NGI was operated at a flow rate of 45 LPM using a downstream vacuum source. As described above, a positive pressure air source was used to deliver the device actuation flow rate of 10 LPM for a time period of 4.6 s resulting in 35 LPM of co-flow air during actuation. The device was actuated with a single 750 ml air volume.

Figure 5.

Profile view of the pediatric air-jet DPI connected to a co-flow adaptor followed by the preseparator inlet of a vertically oriented NGI.

Drug mass depositions in the air-jet DPI and MP together with the NGI impactor components (preseparator, impaction plates and the filter) were determined by washing with deionized water and analysis using the tobramycin LC-MS method (31). The mass of drug deposited in the air-jet DPI and MP was expressed as a % of the nominal loaded dose. The emitted dose (ED) was calculated as loaded tobramycin dose minus tobramycin dose deposited on the DPI.

Stage cut-off diameters were calculated using the method described in USP 35 (Chapter 601, Apparatus 5) for a flow rate of 45 LPM. Each aerosol was characterized by the MMAD, the fine particle fraction < 5µm (FPF_<5µm) and the sub-micrometer FPF (FPF_<1µm), which were expressed as a percentage of the ED. These values were obtained using linear interpolation on a plot of cumulative percentage drug mass vs. cut-off diameter.

The loaded dose of tobramycin was determined by assaying the drug content of a known mass of Tobi-EEG formulation and calculating the mean amount of tobramycin per mg of formulation.

Experimental: Dynamic Vapor Sorption (DVS)

Dynamic vapor sorption experiments (DVS Adventure, Surface Measurement Systems Ltd., UK) were used to evaluate the moisture uptake of some of the new powders based on the methods described in Hassan et al. (31). Briefly, sample mass change of fully dried powders was determined automatically for RH increasing from 0 to 90% at 2% RH/hr. These results were analyzed to estimate the glass transition RH based on observed inflections in the absorption profiles, for different spray drying conditions.

Study Design Overview

In order to determine the best drying conditions for the spray-dried Tobi-EEG formulation using a combination of CFD and experimental methods, a staged iterative approach was adopted. This approach sought to implement CFD to define beneficial reduced drying rate conditions, based on the association between reduced drying rate parameters and improved powder performance for small-particle aerosols as identified by Longest et al. (14). The CFD simulations further sought to predict when, with reduced drying rates, RH conditions in the collector were excessively high, likely degrading powder aerosol performance. Additional spray-dried powder batches were created based on the CFD model to develop enhanced correlations, test the effects of excess RH exposure and verify best case powder performance. The key analysis stages for conducting this study were as follows:

Stage 1: Begin with the previous best-case formulation and spray drying conditions established Hassan et al. (31) (Case 5) and conduct CFD screening simulations to better determine the effects of inlet temperature, inlet water vapor, and inlet flow rate on droplet drying parameters. From the CFD results, select multiple sets of spray drying conditions (cases) for powder production and aerosol testing.

Stage 2: Based on the CFD selections in Stage 1, experimentally produce multiple new powder cases and test aerosolization performance. Use these experimental results to develop quantitative relations between CFD-predicted drying parameters and aerosol performance metrics (FPF). Implement the new experimental results to better understand the effect of excessive RH in the collection region of the spray dryer.

Stage 3: Based in insights gained in Stage 2, implement the CFD model to test additional spray drying conditions and recommend a new lead set of cases.

Stage 4: Based on the CFD findings in Stage 3, produce powders for the lead set of cases and evaluate aerosolization behavior. Compare results with the established CFD correlations. For an established new best case, test lung delivery efficiency and compare with the lung delivery efficiency of the original Case 5 formulation from Hassan et al. (31).

Initially Targeted Spray Drying Conditions

Improved powder performance relative to Case 5 from Hassan et al. (31) is expected for spray drying conditions that achieve:

1. Reduced drying rates (quantified by CFD-predicted κ_avg and κ_max)

2. Collector Surface RH ≤ Maximum Surface RH limit

3. Fully-dried particles at the powder collector (electrostatic precipitator) point of entry

Considering the drying rate, Longest et al. (14) previously demonstrated that with small-particle spray drying in an albuterol sulfate (AS) –EEG formulation, reduced κ_avg and κ_max values can reduce aerosol size and increase emitted dose with vibrating capsule-based DPIs. Benchmark κ_avg and κ_max values will therefore be determined for the Case 5 spray drying conditions from Hassan et al. (31) in the Results. Additional cases will then be sought that achieve lower evaporation rate values.

Using CFD predictions, a Collector Surface RH (%) will be established as the area-weighted average RH on the electrostatic precipitator surface where the powder is collected (outer wall). For a given particle formulation, a Maximum Surface RH (%) limit can likely be established above which the powder aerosol performance is affected by high levels of water uptake, particle-to-particle bridge formation, damage to internal particle structures, increased surface cohesion and possibly glass transition. From spray drying experience, the Max Surface RH for the Tobi-EEG formulation with l-leucine is initially estimated to be 30%. The relationship between the Collector Surface RH and Maximum Surface RH can then be captured as

$$\mathrm{\Delta }{\text{RH}}_{30}=\text{Collector Surface RH}\left(\%\right)–\text{Maximum Surface RH}\left(\%\right)$$ (14)

As a result, negative or zero values of ∆RH₃₀ are desirable and indicate the Collector Surface RH is below the Maximum Surface RH. In contrast, positive values indicate that the Collector Surface RH is excessively high relative to the established Maximum Surface RH, which will likely degrade powder aerosol performance. In cases of positive ∆RH₃₀, decreased emitted dose is initially expected.

Finally, while reduced drying rates (14) and increased collector RH (31) may be desirable up to a limit, spray drying conditions achieving these goals may be detrimental to powder formation and performance if the particles are not fully dry at the point of entry into the collector. Initial particle dryness can be estimated using the CFD-calculated mass fraction of liquid in each particle (mf_liquid) ≤ 0.001 w/w (or ≤0.1% w/w). Viewed from the perspective of solutes, it is desirable to have mf_solute > 0.999 w/w before collector entry. As with the other initial spray drying targets, these initial estimates will be compared with the experimental evidence and adjusted as needed.

Statistical Analysis

Replicate analysis from at least three measurements were used to calculate mean and standard deviation (SD). Statistical differences were tested using Student’s t-test, one-way ANOVA followed by Tukey’s HSD (p-value < 0.05) or a multivariance standard least square fitting model. Evaluations were performed using JMP Pro (SAS Institute Inc., Cary, NC).

RESULTS

Stage 1 Analysis

Initial (Case 5) and other spray drying conditions evaluated in Stage 1 are summarized in Table 1. Spray drying conditions explored reduced inlet drying temperature below 70 °C, varied the inlet water vapor content (10, 5 and 1 g/m³) and increased drying gas flow rate to improve downstream RH conditions. As shown in Table 1 for Case 5, the CFD-predicted Collector Surface RH was above the initially selected Maximum Surface RH of 30% resulting in a ∆RH₃₀ value of 8.6%. Case 5a conditions significantly reduced the ∆RH₃₀ value to −19.0% while maintaining κ_avg and κ_max values approximately consistent with Case 5. All other cases achieved reduced evaporation rates, which are expected to improve aerosolization performance. However, only some of the remaining cases satisfied the negative ∆RH₃₀ requirement. Of the Stage 1 cases, 5h appears to provide a best combination of spray drying conditions with ∆RH₃₀ = −8.3% and the lowest Stage 1 κ_avg and κ_max values.

Table I.

Stage 1 Cases Evaluated with CFD

Case	Inlet Temp. (°C)	Inlet water vapor (g/m3)	Inlet flow rate (LPM)	Collector Temp. (°C)	Collector surface RH (%)	∆RH (%)	Droplet κavg (µm2/ms)	Droplet κmax (µm2/ms)
5	70	10	120	32.8	38.6b	8.6	5.2 (68.2)a	155 (70.7)a
5a	60	1	120	30.1	11.0	−19.0c	5.2 (69.4)	149 (80.0)
5b	60	10	120	30.0	44.0b	14.0	4.0 (72.8)	124 (78.6)
5c	55	1	120	28.7	11.7	−18.3c	4.7 (69.4)	136 (80.1)
5d	55	10	120	28.5	47.0b	17.0	3.3 (72.5)	115 (64.5)
5e	55	1	160	30.1	6.0	−24.0c	4.4 (67.9)	137 (76.1)
5f	55	10	160	30.0	38.0b	8.0	3.1 (77.1)	118 (94.5)
5g	50	1	160	28.6	6.4	−23.6c	3.6 (77.2)	124 (83.2)
5h	50	5	160	28.6	21.7	−8.3c	3.0 (74.2)	114 (68.7)

^aMean values averaged over all representative droplets (coefficient of variation)

^bCollector Surface RH value greater than Maximum Surface RH (30%)

^cNegative values indicate that the Collector Surface RH is below the Maximum Surface RH (30%)

Additional details regarding the Stage 1 spray drying conditions can be observed from the plotted heat and mass transfer profiles (Figures 6 and 7). Considering midplane temperature profiles in the spray dryer (Figure 6), significant cooling from the wall boundary is observed across the multiple selected cases. As a result, outlet temperatures are relatively similar among the cases. Specifically, inlet temperatures ranged from 50 to 70 °C (∆T = 20 °C) and produced outlet temperatures ranging from 28.6 to 32.8 °C (∆T = 4.2 °C). It is also observed that due to wall cooling, the collected powder is exposed to lower temperatures on the outer wall of the collector region, which may help prevent powder degradation from prolonged high temperature exposure during the spray-drying.

Figure 6.

CFD-predicted midplane and cross-sectional temperature profiles for: (a) Case 5, (b) Case 5a, (c) Case 5b, (d) Case 5e and (e) Case 5h. Values of T_collector represent the area-averaged temperature value on the surface of the electrostatic precipitator where the powder is collected.

Figure 7.

CFD-predicted midplane and cross-sectional relative humidity (RH) profiles for: (a) Case 5, (b) Case 5a, (c) Case 5b, (d) Case 5e and (e) Case 5h. Values of RH_collector represent the area-averaged relative value on the surface of the electrostatic precipitator where the powder is collected (Collector Surface RH).

Considering RH profiles, the reduced temperature from the wall cooling increases RH values on the powder collector surface (Figure 7), which may, in contrast with lower temperature, affect powder aerosol performance. Case 5b in Figure 7 displays a high Collector Surface RH value of 44%, which is significantly above the Maximum Surface RH of 30% (∆RH₃₀ = −8.3%).

CFD-predicted particle conditions and drying parameters are depicted in Figure 8 for Case 5 conditions. Even with the relatively high RH conditions of Case 5, evaporation is observed to occur rapidly (Figure 8a) with mf_solute values > 0.999 occurring before half the drying chamber length (Figure 8b), indicating nearly fully dry particle conditions. As described by Longest et al. (14) for a different formulation, high variability is observed in evaporation rate parameters κ_avg (Coefficient of Variation among different droplets; CoV = 68.2%) and κ_max (CoV = 70.7%).

Figure 8.

CFD-predicted droplet characteristic and drying parameters for Case 5 conditions including (a) droplet diameter, (b) mass fraction of total solutes (w/w) in the droplet (mf_solute), (c) average drying rate (κ_avg), and (d) maximum drying rate (κ_max).

Based on CFD predictions, cases selected for spray drying production and testing in Stage 2 are 5, 5a, 5b, 5e and 5h. These selections are intended to capture a range of drying evaporation values and Collector Surface RH conditions. Based on the CFD predictions, Cases 5 and 5b are expected to have excess values of ∆RH₃₀. In contrast, Cases 5a, 5e and 5h have decreasing drying rate values and satisfy the Collector Surface RH criterion. Cases 5a and 5e have very low Collector Surface RH values, whereas 5h is nearest to and below 30%, with an ∆RH₃₀ = −8.3%.

Stage 2 Analysis

Powders for the five selected cases from Stage 1 were produced using the Buchi Nano B-90 HP Spray Dryer. As described in the Methods, the powders were collected and then stored for a 1 week relaxation period. Powders were then loaded into the pediatric air-jet DPI and tested for aerosol performance using cascade impaction. As observed in Table 2, Cases 5 and 5b (which had excess Collector Surface RH values) were highly similar and displayed reduced ED (67%) but also low preseparator deposition fraction (<5%) and high FPF_<5µm (>89%).

Table II.

Stage 2 Experimental Testing (Cascade Impaction) of Selected Cases

Description	Case 5	Case 5a	Case 5b	Case 5e	Case 5h
ED (%)*	67.0 (3.0)	72.0 (1.6)**	66.9 (3.0)	73.3 (3.5)**	76.1 (2.2)**
Preseparator (%)*	3.8 (0.5)	33.6 (1.8)**	4.6 (0.3)	28.7 (1.9)**	11.7 (2.8)**
FPF<5µm (%)*	89.4 (2.0)	40.9 (2.5)**	88.7 (1.2)	49.6 (2.6)**	75.4 (6.7)**
FPF<1µm (%)*	14.8 (2.3)	7.6 (0.3)**	15.9 (2.5)	8.8 (0.7)**	18.4 (2.2)**
MMAD (µm)*	1.8 (0.1)	1.9 (0.1)	1.9 (0.1)**	2.0 (0.1)**	1.7 (0.1)

^*p<0.05 significant effect of case on ED, Preseparator, FPF_<5µm, FPF_<1µm, and MMAD (one-way ANOVA).

^**p<0.05 significant difference compared to Case 5 (post-hoc Tukey).

Mean values with standard deviations (SD) shown in parenthesis [n=3].

Cases with Collector Surface RH < 30% (5a, 5e and 5h) had higher ED values compared with 5 and 5b, and the ED values of these cases were similar to each other (≥ 72%). These Collector Surface RH < 30% cases also had increasing FPF_<5µm that appears to correspond with decreasing drying rates (κ_avg and κ_max). Furthermore, preseparator deposition fraction was excessively high (~30%) for the cases with much lower Collector Surface RH values (Cases 5a and 5e). As Collector Surface RH increased to 21% with Case 5h, the preseparator deposition fraction decreased to 11%. To summarize these experimental observations:

• Collector Surface RH > 30% reduced ED;

• reducing κ_avg and κ_max improved FPF;

• very low Collector Surface RH (<20%) dramatically increased preseparator deposition fraction (indicating possibly high particle aggregation); and

• achieving Collector Surface RH in a range of 20–30% with low κ_avg and κ_max produced the best overall powder performance (Case 5h).

CFD-predicted drying parameters, established in Longest et al. (14), are plotted against FPF_<5µm in Figure 9. Strong linear correlations between FPF_<5µm and three of the drying parameters are observed for the cases with Collector Surface RH < 30% (i.e., Cases 5a, 5e and 5h). These correlations occur for κ_avg²d_final, κ_max²d_final and ISR²d_final with all R² values ≥ 0.97. In contrast, Cases 5 and 5b (with excessive Collector Surface RH) do not follow the expected trend and behave in an unpredictable way, but with higher FPF_<5µm values. A possible explanation is that the larger powder particles are disproportionately affected by high RH exposure becoming more cohesive and causing the reduced ED observed in Table 2. With reduced ED due to retention of the larger particles in the inhaler, the aerosol that does escape the aerosolization chamber of the air-jet DPI has a smaller FPF.

Figure 9.

Experimentally determined aerosol fine particle fraction <5 µm (FPF_<5µm) vs. drying parameters (a) κ_avg²d_final, (b) κ_max²d_final, (c) ISR²d_final, and (d) precipitation window t_p.

Stage 3 Analysis

Based on observations from Stage 2, conditions were selected to further reduce evaporation rate and maintain the Collector Surface RH very near 30%. These conditions were achieved by further reducing drying temperature while also modifying the water vapor content and drying gas flow rate (Table 3). CFD predictions of these three new cases indicate a Collector Surface RH range of 26.7 to 32.7%. Cases with a 35 °C drying temperature (5j and 5k) provide the lowest drying rate parameters observed. Cases 5j and 5l have Collector Surface RH values just above the 30% limit. Case 5k has a Collector Surface RH just below the limit (∆RH₃₀ = −3.3%) with the lowest observed κ_avg value of 1.8 µm²/ms.

Table III.

Stage 3 Cases Evaluated with CFD

Case	Inlet Temp. (°C)	Inlet water vapor (g/m3)	Inlet flow rate (LPM)	Collector Temp. (°C)	Collector surface RH (%)	∆RH (%)	Droplet κavg (µm2/ms)	Droplet κmax (µm2/ms)
5	70	10	120	32.8	38.6b	8.6	5.2 (68.2)a	155 (70.7)a
5h	50	5	160	28.6	21.7	−8.3c	3.0 (74.2)	114 (68.7)
5j	35	1	80	20.9	32.7b	2.7	2.2 (83.6)	81 (92.4)
5k	35	5	160	24.3	26.7	−3.3c	1.8 (85.2)	81 (61.7)
5l	50	7	145	28.6	31.8b	1.8	2.8 (80.8)	110 (91.1)

^aMean values averaged over all representative droplets (coefficient of variation)

^bCollector Surface RH value greater than Maximum Surface RH (30%)

^cNegative values indicate that the Collector Surface RH is below the Maximum Surface RH (30%)

As illustrated in Figure 10, the low inlet drying temperature of Cases 5j and 5k (35 °C) results in very low outlet temperatures of 20.9 and 24.3 °C, respectively, which may be insufficient in drying the aerosol. These low drying temperatures resulted in high central core RH values extending all the way to the electrostatic precipitator. It is expected that this elevated RH may substantially reduce particle drying throughout the drying chamber. Figure 11 illustrates the solute mass fraction (mf_solute) for 150 randomly selected particle trajectories with Cases 5j, 5k and 5l. As expected for Case 5k, particles with mf_solute < 0.999 (not fully dry) are shown to enter the electrostatic precipitator. Cases 5j and 5l do not violate this condition.

Figure 10.

CFD-predicted midplane and cross-sectional temperature profiles for: (a) Case 5j, (b) Case 5k, and (c) Case 5l. Corresponding relative humidity (RH) profiles for (d) Case 5j, (e) Case 5k, and (f) Case 5l. Values of T_collector and RH_collector represent the area-averaged temperature and relative humidity values on the surface of the electrostatic precipitator where the powder is collected.

Figure 11.

Mass fraction of total solutes for individual droplets from approximately 150 randomly sampled trajectories for (a) Case 5j, (b) Case 5k and (c) Case 5l. Case 5k was determined to have insufficient drying due to droplets with mf_solute ≤ 0.999 reaching the collector region.

Based on these observations, none of the three new cases satisfied both recommended conditions of Collector Surface RH < 30% and near fully dried particles reaching the collection region. Case 5l was assumed to be the best option based on satisfying the fully dry particle criteria and having a Collector Surface ∆RH₃₀ of only 1.8%. Nevertheless, all three powders were very close to satisfying the recommended drying conditions and the exact values for the best drying limits require additional refinement. Therefore, all three powders were selected for production and testing in Stage 4.

Stage 4 Analysis

The selected three new spray drying conditions were implemented and the resulting powders were aerosolized with the pediatric air-jet DPI (Table 4) after the 1-week powder relaxation period. From Table 4 it is observed that the incomplete drying of Case 5k results in a sizable reduction in FPF_<5µm (<70%) and subsequent increase in preseparator fraction (>10%). As a result, Case 5k is excluded from further consideration. Case 5l maintained a sufficiently high ED similar to the best case from the previous stages (Case 5h). FPF_<5µm was also increased with Case 5l relative to Case 5h and similar to Case 5. Considering Case 5j, with a slightly higher ∆RH₃₀ = 2.7%, the device ED begins to decrease. Trends in these experimental results support the conclusions that:

Table IV.

Stage 4 Experimental Testing (Cascade Impaction) of Selected Cases

Description	Case 5	Case 5j	Case 5k	Case 5l
ED (%)*	67.0 (3.0)	71.9 (3.1)	72.6 (3.5)	74.1 (2.2)**
Preseparator (%)*	3.8 (0.5)	9.5 (1.3)**	15.3 (3.6)**	7.4 (0.6)
FPF<5µm (%)*	89.4 (2.0)	84.0 (2.2)	69.4 (5.3)**	86.4 (1.6)
FPF<1µm (%)	14.8 (2.3)	15.2 (0.4)	12.8 (3.0)	15.9 (0.7)
MMAD (µm)	1.8 (0.1)	1.7 (0.1)	1.8 (0.1)	1.7 (0.0)

^*p<0.05 significant effect of case on ED, Preseparator, and FPF_<5µm (one-way ANOVA).

^**p<0.05 significant difference compared to Case 5 (post-hoc Tukey).

Mean values with standard deviations (SD) shown in parenthesis [n=3].

• particles should satisfy the mf_solute > 0.999 condition prior to collector entry (or FPF is reduced, as with Case 5k);

• the Collector Surface RH should remain below approximately 32% (as with Case 5l); and

• Collector Surface RH values marginally above 32% start to reduce ED (as with Case 5j).

The effect of adding Case 5l to the comparison of FPF vs. CFD-predicted drying parameters is illustrated in Figure 12. In these comparisons, only cases that achieve a Collector Surface RH <32% are included (5a, 5e, 5h and 5l). It is noted that Cases 5l and 5h achieve the highest and second highest FPFs, respectively. Overall, strong correlations are observed for three of the drying parameters compared with FPF using either a linear or second order polynomial fit. Drying parameters displaying strong correlations (R² = 1.0) are κ_avg²d_final, κ_max²d_final and ISR²d_final. A correlation is not observed between FPF and precipitation window, t_p (Figure 12d). These strong correlations indicate that aerosol performance of spray-dried powders from an air-jet DPI is highly predictable based on CFD simulations provided that Surface Collector RH is maintained below a specific level, which for Tobi-EEG with l-leucine appears to be approximately 30% based on Figure 12.

Figure 12.

Experimentally determined aerosol fine particle fraction <5 µm (FPF_<5µm) vs. drying parameters (a) κ_avg²d_final, (b) κ_max²d_final, (c) ISR²d_final, and (d) precipitation window t_p. Only cases that satisfy the recommended spray drying conditions for Collector Surface RH and mf_solute are included.

Based on aerosolization testing, Case 5l conditions appear to produce the best overall Tobi-EEG l-leucine powder. Compared with the initial Cases 5 powder, Case 5l demonstrates statistically significant increases in ED and FPF_<1µm. However, the preseparator deposition fraction and FPF_<5µm indicated statistically significant decreases in performance. The MMAD of Case 5l did trend lower to a value of 1.7 µm, but this small decrease was not significant compared with a Case 5 value of 1.8 µm.

Based on best case aerosolization performance, Case 5l powder was delivered to the pediatric MT airway model and compared with Case 5 conditions (Table 5). As expected, the increased ED of Case 5l produced a statistically significant increase in tracheal filter delivery from 55.2% to 62.9%. However, the reduced FPF and increased preseparator deposition fraction of Case 5l doubled the MT deposition to a value of 18.4% compared with Case 5. Hence, Case 5l indicated improved performance with maximum lung delivery, but at the expense of increased MT depositional loss.

Table V.

Stage 4 Comparison of Initial (Case 5) and lead (Case 5l) Conditions with the Pediatric Mouth-Throat (MT) Model

Description	Case 5	Case 5l
Device (%)*	21.7 (2.6)	7.9 (1.6)
MP (%)	12.7 (1.3)	13.8 (0.8)
ED (%)*	65.6 (1.4)	78.4 (0.9)
MT (%)*	9.3 (1.5)	18.6 (0.8)
Filter (%)*	55.2 (0.8)	62.9 (2.2)

^*p <0.05 significant difference (t-test).

Mean values with standard deviations (SD) shown in parenthesis [n=3].

Data Summary

Cases for which both CFD results and experimental data are available from this study are plotted in Figure 13 as FPF_<5µm vs. ED. Each data point is identified and grouped according to having sufficiently low RH (defined as RH ≤ 30%), high RH or low evaporation (violating either the RH or fully evaporated particle criteria), or an intermediate case. In this figure, Case 5l was termed an intermediate case based on violating the assumption of <30% Collector Surface RH but still satisfying a limit of <32% Collector Surface RH. Interestingly, cases with sufficiently low RH form a linear correlation in which increasing ED is also associated with increasing FPF. Furthermore, decreasing κ_avg values are associated with moving up this curve and improving aerosol quality and performance in terms of higher FPF and higher ED. Case 5l, with ∆RH₃₀ = 1.8%, is close to but falls just outside this curve due to slightly reduced ED, presumably because the Collector Surface RH is just above the 30% limit. Case 5k is not on the curve, presumably because of violating the fully dried particle requirement. All other cases demonstrate more substantially reduce ED based on high ∆RH₃₀ values. Moreover, excluding Case 5k with incomplete evaporation, increasing values of ∆RH₃₀ (indicating excessive humidity) appear to correlate with decreased ED.

Figure 13.

Experimentally determined aerosol fine particle fraction <5 µm (FPF_<5µm) vs. experimentally determined emitted dose (ED) for all cases with available data. Cases that satisfy the recommended spray drying conditions related to Collector Surface RH and mf_solute fall along a line where reducing κ_avg improves both FPF and ED.

DISCUSSION

Best-case spray drying conditions (Case 5l) were identified that provided a significantly improved tracheal filter delivery efficiency in a pediatric MT model reaching a value >60% of inhaler loaded dose. While this increase was less than expected, the more significant outcome of this study was an improved understanding of the relationship between small-particle spray drying conditions and powder performance. These insights were made possible through an application of both CFD simulations and in vitro experiments in an iterative manner. Assimilating the results, key findings indicate:

• reduced drying rate parameters (κ_avg and κ_max) have a beneficial effect on the aerosolization of powders produced from a small-particle spray dryer;

• the Collector Surface RH should be maintained within a range, which for the Tobi-EEG formulation with l-leucine was determined to be approximately 20–30%; and

• CFD simulations should ensure near full drying of the particles, which in this study was found to be captured with mf_solute >0.999.

These new small-particle spray drying recommendations are explored further below.

A surprising finding in small-particle spray drying appears to be improving aerosol performance with conditions that reduce CFD-predicted κ_avg and κ_max. This finding is in contrast with most other published studies on predicting the performance of spray-dried powders (1, 2, 18), which advocate for more rapid drying. It is noted that ours is the only work to predict evaporation rates with CFD across all droplets in the spray drying system, as opposed to predictions based on the drying rate of a single representative droplet. Evidence of improved performance between lower κ_avg (or κ_max) and FPF is clearly apparent in Figures 9 and 12 with coefficients of determination of approximately R² ≈ 1. Interestingly, this association was only true for cases with Collector Surface RH below a value of approximately 30%. Moreover, Figure 12 indicates that both FPF and ED improve with deceasing κ_avg when using an air-jet DPI for aerosolization. The use of the air-jet DPI may account for this unexpected relationship; however, similar findings were observed in Longest et al. (14) when using a vibrating capsule-based DPI. As further described in Longest et al. (14), we expect the relationship between reduced evaporation rate and improved powder aerosolization performance to be unique to small droplet/particle spray drying as achieved with the Buchi B-90 HP. When starting with initial droplets of approximately 6 µm, sufficient time may be required for shell or porous particle formation. Furthermore, while 6 µm droplets may have time for porous particle formation, smaller droplets in the polydisperse size distribution may not have sufficient time leading to the need for slower evaporation rates to improve the overall performance of the powder across a polydisperse size range. While results show that increased drying time is an important factor, it was surprising that powder performance did not correlate with the precipitation window parameter, t_p, in this study.

At the study outset, it was anticipated that there would be an upper limit to Collector Surface RH, and exposure beyond this limit would affect powder aerosolization performance. Attempts to base this limit on the variable glass transition RH as determined by Hassan et al. (31) were not successful in preliminary testing. Based on previous experimental experience, we initially anticipated that a Maximum Surface RH of 30% would be an appropriate first estimate. Results of Figure 9 showed that this estimate was reasonable based on association between drying parameters and FPF. In exploring Collector Surface RH values closer to 30% in Stages 2 and 3, it appeared that 32% may be an acceptable upper limit based on Case 5l falling along the correlation curve. However, plotting all experimental data in Figure 13 highlighted that ED with Case 5l was just beginning to decrease relative to the other cases causing us to select 30% as the likely best Maximum Surface RH for the Tobi-EEG l-leucine formulation.

Considering the lower limit of Collector Surface RH, the previous study of Hassan et al. (31) established that glass transition RH was increased for powders sprayed with increased drying gas water vapor content, which indicates improved powder stability. However, aerosolization performance of both low and high water vapor content powders was not compared using an air-jet DPI. In the current study it was surprising that overly dry conditions on the collector surface (Collector Surface RH <20%) produced very high preseparator deposition fractions and low FPFs. It is noted that FPF calculations take into account aerosol loss in the preseparator whereas MMAD calculations using a standard algorithm do not. The high preseparator mass fraction of drug would be lost in the MT during in vitro testing or in vivo usage. Based on experimental observations, the two powders with the lowest Collector Surface RH (Cases 5a and 5e) appeared highly charged during device loading, even after the 1 week relaxation period. It is reasonable that higher RH with other cases helps to discharge the powder in the collector through the action of charged water vapor ions as described by Nguyen and Nieh (51). This phenomenon needs further exploration and may be a source of improved powder performance.

Considering the CFD-predicted drying rate parameters further, Tables 1 and 3 provide valuable information to select targeted values. From Stages 1 and 2 of the analysis (Table 1), reducing the drying rate κ_avg and κ_max parameters to approximately 3 and 114 µm²/ms significantly improved aerosol performance among cases satisfying the RH and mf_solute conditions (Table 2 and Figure 9). Based on Stages 3 and 4 (Table 3), reducing the drying rate parameters below these values to 2.8 and 110 µm²/ms further improved powder performance (Table 4 and Figure 12). As a result, it can be concluded that targeted drying rate parameters for future studies with small-particle spray drying should be κ_avg and κ_max values below 3 and 114 µm²/ms, respectively, while also satisfying the RH and mf_solute criteria.

As observed by Hassan et al. (31), a concern with overly rapid drying of droplets may be low glass transition RH values, indicating less stable powders. Recommended drying conditions should take both aerosolization and stability of the resulting powder into consideration (52). In Figure 14 glass transition RH values are plotted from Hassan et al. (31) plus three additional lead cases from this study based on DVS measurements vs. κ_avg and κ_max values. Interestingly, glass transition RH values correlate well with the CFD-predicted drying parameters. As with aerosolization behavior, κ_avg and κ_max values below 3 and 114 µm²/ms are associated with the highest glass transition RH values in the range of >40%. In these figures a zone of Best Performance is also shown highlighting the recommended RH range for Surface Collector RH together with limits on κ_avg and κ_max. As illustrated, meeting the three small particle spray drying recommendations in this study is associated with high quality aerosolization and increased glass transition RH indicating improved powder stability.

Figure 14.

Experimentally determined Glass Transition RH vs. CFD predicted (a) κ_avg and (b) κ_max. Low values of κ_avg and κ_max are observed to produce beneficial high (>40%) Glass Transition RH values.

Based on results of this study, the spray drying conditions of Case 5l may be the best possible for the Tobi-EEG formulation spray-dried in the Buchi B-90 HP. If MT deposition needs to remain below 10%, then Case 5 conditions established by Hassan et al. (31) may be best. Considering further improvements beyond Case 5l, a challenge is that changes to reduce the drying rates typically increase Collector Surface RH. This partially arises from the significant cooling that is occurring within the spray drying chamber and wall heat loss (Figures 6 and 10). Attempts to reduce the marginally high ∆RH₃₀ of Case 5l will likely increase the κ_avg value (2.8 µm²/ms) to that of Case 5h (3.0 µm²/ms; see Figure 13), resulting in no net improvement in powder performance. Therefore, it may not be possible to further improve the performance of Case 5l based on the recommended spray drying conditions developed in this study. Potential options to further improve powder performance while maintaining the conditions identified in this study include: (i) modifying the spray dryer design to allow for targeted κ_avg and κ_max conditions while also maintaining Collector Surface RH in the 20–30% range; (ii) changing powder formulations with a focus on the dispersion enhancer to better ensure the relatively high RH values that appear to be beneficial to powder formation and performance; and (iii) finding better ways to remove charge from the powder after low RH drying or to collect the powder with less exposure to charge formation.

Limitations of this study can be grouped into the categories of CFD simulations, in vitro testing and study outcomes. Considering the CFD model, steady state simulations were performed to enable the inclusion of two-way coupling of the heat and mass transfer fields in a time efficient manner. With these assumptions, simulations required 2–3 days per spray dryer case on a high-performance desktop machine. Inclusion of transient simulations would increase this time scale by approximately an order of magnitude. While steady state simulations offered valuable insights, transient oscillations of the nebulizer jet and droplet flow field are often observed in actual spray dryer operation. Furthermore, oscillations of flow in the collector and around the highly simplified charged electrode and small diameter spray dryer outlet orifice can also occur causing shifts in the Collector Surface RH, especially in regions of high RH gradients (as observed in Figure 7c in the collection region).

Considering spray drying and in vitro aerosolization testing, only the Buchi B-90 HP spray dryer was implemented with a medium mesh configuration. The relationships identified between evaporation rate and aerosolization performance may not be valid for smaller or larger initial droplets. Aerosol deposition in the pediatric MT model has not been validated with in vivo data; however, the pediatric MT model has key geometric features that are very similar to in vivo measurements (53, 54). Furthermore, our intent was to evaluate the performance of the spray-dried powders when aerosolized with a pediatric air-jet DPI. As illustrated in other studies, design parameters of the air-jet aerosolization engine (including inlet and outlet flow passages and the aerosolization chamber) can significantly impact performance (55–60). The selected DPI design was intended to serve as a common testing benchmark in this study. Performance of the low Collector Surface RH powders may be substantially improved with a different aerosolization engine design.

Quantitative study outcomes may be specific to the spray dryer, air-jet DPI and formulation implemented. It is expected that the three general recommendations arising from this study are applicable for most small-particle spray drying cases and air-jet DPI devices. However, the range of Collector Surface RH values and mf_solute limit likely vary based on the selected powder formulation. Limitations of the Buchi B-90 HP spray dryer identified in this study can likely be overcome using formulations that are more resistant to higher RH exposure.

Conclusions

In conclusion, improved aerosolization performance of the powder was achieved by finding an optimal balance between droplet evaporation rate and collector surface relative humidity conditions. For the small-particle spray drying conditions considered, reducing the droplet evaporation rate produced improved aerosolization performance with increased FPF and ED. As the droplet evaporation rate values were lowered, it was important to not exceed a collector surface relative humidity maximum condition, which was shown to significantly decrease the ED of the aerosolized powder. There was also a substantial benefit to maintaining the Collector Surface RH condition within an elevated range. Comparing CFD predictions with limited spray-dried powder batches and experimental testing was demonstrated to be a useful technique to identify the optimal balance between droplet evaporation rate and Collector Surface RH conditions. CFD was also useful to ensure that the particles entering the collector were nearly fully dry. For the specific Tobi-EEG formulation considered in this study, best aerosolization performance was achieved by (i) reducing the CFD-predicted drying parameters of κ_avg and κ_max to values below 3 µm²/ms and 114 µm²/ms, respectively, (ii) maintaining the Collector Surface RH within an elevated range, which for the Tobi-EEG formulation with l-leucine was 20–30 %RH; and (iii) ensuring that particles reaching the collector were fully dried, based on a mass fraction of solute CFD parameter. By implementing these guidelines, improved powder aerosolization performance was achieved that increased in vitro determined pediatric lung dose beyond the initially targeted value of 60% based on inhaler loaded dose. Further improvements in aerosol performance likely cannot be achieved through changes in spray drying conditions alone and may require changes to spray dryer design, powder formulation and post production powder processing.

ABBREVIATIONS

3D

three dimensional

AS

albuterol sulfate

CFD

computational fluid dynamics

CoV

coefficient of variation

DPI

dry powder inhaler

DVS

dynamic vapor sorption

ED

emitted dose

EEG

excipient enhanced growth

FPF

fine particle fraction

HP

high performance

HPLC

high performance liquid chromatography

ISR

time integral of saturation ratio

LC-MS

liquid chromatography-mass spectrometry

LPM

Liters per minute

LRN

low Reynolds number

MMAD

mass median aerodynamic diameter

MN

mannitol

MP

mouthpiece

MT

mouth-throat

NGI

Next Generation Impactor

ODE

ordinary differential equation

PDE

partial differential equation

RH

relative humidity

RMM

rapid mixing model

SD

standard deviation

SR

saturation ratio

T

temperature

Tobi

tobramycin

USP

United States Pharmacopeia