If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Long-acting implants are typically formulated using carrier(s) with specific physical and chemical properties, along with the active pharmaceutical ingredient (API), to achieve the desired daily exposure for the target duration of action. In characterizing such formulations, real-time in-vitro and in-vivo experiments that are typically used to characterize implants are lengthy, costly, and labor intensive as these implants are designed to be long acting. A novel characterization technique, combining high resolution three-dimensional X-Ray microscopy imaging, image-based quantification, and transport simulation, has been employed to provide a mechanistic understanding of formulation and process impact on the microstructures and performance of a polymer-based implant. Artificial intelligence-based image segmentation and image data analytics were used to convert morphological features visualized at high resolution into numerical microstructure models. These digital models were then used to calculate key physical parameters governing drug transport in a polymer matrix, including API uniformity, API domain size, and permeability. This powerful new tool has the potential to advance the mechanistic understanding of the interplay between drug-microstructure and performance and accelerate the therapeutic development long-acting implants.
Recently, a considerable amount of research has been conducted on long-acting (LA) drug delivery systems for improving patient compliance and treatment outcomes. These goals are typically achieved by maintaining therapeutic drug concentrations over an extended period of time, thereby minimizing adverse side effects associated with the exposure to high amount of drug and reducing the dosing frequency.
The release rate of these drug products is influenced by the physicochemical properties of the API and the polymer matrix, matrix microporosity if any, as well as the interactions between water, API, and polymer. The microstructures of these drug products are complex in nature, and are often controlled by formulation composition, manufacturing process, and material properties. Characterization of implant microstructure and drug distribution, and their temporal changes, is critical to the understanding of the drug release mechanisms and consequently to guide the development of drug products.
During the development of LA systems, significant time and effort are required for the following activities: (a) In vitro release testing necessary for optimizing the drug product formulation, (b) batch release of drug product, and (c) the development of in vitro-in vivo correlations (IVIVC). Additionally, if accelerated in vitro testing methods are not available for product development and batch release, it may significantly impact the development timelines. Novel characterization tools and analysis approaches, such as high-resolution imaging, advanced imaging analytics, and image-based modeling and simulation emerge as alternatives that can be beneficial for both fundamental understanding and for predicting drug release rates.
However, there were two major hurdles retarding the general application of this powerful tool in pharmaceutical sciences: (i) the ability to obtain high resolution and good contrast images for soft, multiphase pharmaceutical samples composed of materials with low atomic numbers and subtle density differences, and (ii) the management, segmentation, and analysis of enormous amounts of imaging data. Advances in instrumentation and accumulated operator experience over the years have helped to mitigate the first issue, while the recent emergence of artificial intelligence (AI)-based cloud imaging analytics tools
In this manuscript we describe the characterization of a LA drug product, developed as a poly(lactic-co-glycolic acid) (PLGA) implant, with high resolution XRM imaging and image-based analytics. Complex microstructures composed of multiple material phases (API-rich region, polymer-rich region, and microporosity) were revealed non-destructively at high resolution in 3D. AI-based image segmentation and quantitative analysis were conducted on the 3D image data to convert “visualized” structural properties into voxelized numerical drug product models. Quantifications of several implant samples from XRM images were used not only to confirm dimensional consistency and surface roughness for quality control of bulk properties, but also API size distribution, API-polymer dispersion uniformity, and porosity inside the final implant product. The voxel-based representation of the drug dispersion network and porosity network were used to calculate characteristic parameters of drug transport in the polymer matrix, including permeability using a computational fluid dynamics simulation. Our findings suggest that X-ray imaging combined with image-based analytics can substantially accelerate the development of long acting implants.
Materials and Testing Methods
In this study, an active pharmaceutical ingredient (API) was dispersed in PLGA to form a cylindrical implant using a hot melt extrusion process. Characteristic properties of the API are shown in Table 1.
The API is zwitterion with a high-water solubility (>50 mg/mL) above pH 5.5. The solution for spray drying was made by adding 30 grams of the API to endotoxin free water to make a 15 mg/mL suspension with a pH around 3.8. 0.5 M sodium hydroxide solution or 0.5 M ammonium hydroxide solution was gradually added to the suspension until a clear solution was formed at around pH 5.6. The base solution addition was continued till the pH reached 7.4. At this pH the final peptide concentration was around 14.7 mgs. The solution was sterile filtered using a Omnipore hydrophilic PTFE membrane filter prior to spray drying. Hereafter, API particles produced from spray drying from the 0.5 M NaOH solution will be referred to as API-NaOH and while that spray dried from 0.5 M NH4OH solution will be referred to as API-NH4OH. The API-NaOH has residual Na left behind in the solid. Due to the volatile nature of ammonium, there is little to no ammonium left behind in the API-NH4OH. The presence or absence of counterion in the solid has the potential to impact properties such as hygroscopicity and release characteristics of the API from the implant. To understand the influence of counterion on implant performance, implant samples manufactured from both API-NaOH and API-NH4OH were investigated.
Imaging characterization was conducted using XRM on fifteen implants formulated with two PLGA types (85/15 or 75/25, both with molecular weight of 100 kDa) at five drug loads (15%, 20%, 25%, 30%, or 35%) produced at two different extrusion temperatures (90°C or 120°C). The API was found to be chemically stable when subjected to extrusion in the presence of PLGA at 120°C. The desired release duration of the API from the rod was established to be 3–4 months for this long acting product. PLGA polymers with L/G ratio >70% are required to meet the intended release duration. Thus, PLGA polymers 75/25 and 85/15 were chosen for investigation.
All PLGA polymers used in the study were obtained from Evonik Inc. (Birmingham, Alabama). Table 2 summarizes the samples investigated in this study along with relevant information on formulation conditions. Structural properties such as drug domain size and microporosity in the polymer matrix were then extracted from the 3D XRM image data and quantified as a function of drug loading. Direct numerical simulations were conducted on the microstructures reconstructed from the image data to predict transport properties such as permeability.
Table 2Samples investigated in this study and the manufacturing conditions used for these samples.
Preparation of the Long Acting Implant Via a Hot melt Extrusion Process
The PLGA-based implants were produced using a hot melt extrusion process at two different sites are summarized in Table 2. Samples in rows 1–8 were manufactured at facility A (designated as facility A samples hereafter), while samples in row 9–15 were manufactured at facility B (designated as facility B samples here after). The facilities are differentiated by shading in Table 2. A green separation line is also added in Tables 2, 4 and Fig. 3 to further distinguish samples made in the different facilities. In both facilities the extrusion of the implant samples was carried out at 90–120 °C using a Pharmamini system (Thermo Fischer Scientific) equipped with conical twin screws operating in the counter rotating mode.
For the facility A samples, the API at appropriate loading was manually blended with PLGA and then extruded at 40 RPM screw speed using a 0.5 mm die. A 30 min recirculation time was used prior to extrusion to ensure uniform blending of the API and PLGA. The 0.5 mm diameter strands from the extruder were cut manually to a length of 5 mm.
For facility B samples, the API at appropriate loading was blended with PLGA in an Acoustic Mixer (Resodyn Inc.) at 40% intensity for 4 min. A recirculation time of 10 min was used prior to extrusion to ensure uniform blending of the API and PLGA. The screw speed was set to 55 RPM during recirculation period and to around 8–10 RPM during extrusion. A die diameter of 0.65 mm was used. The strands from the extruder were cut using a Stericut system (Thermo Fischer Scientific) to a length of 5 mm. The cut segments of the implants were analyzed by RP-UPLC for drug content.
API Quantitation by RP-UPLC
API content in drug product formulations was determined using Agilent UHPLC system equipped with a binary pump. 5 µL of sample solution in 30/70 THF/water (v/v) was injected on to a Waters XBridge BEH C4 column (50 mm x 4.6 mm i.d., 3.5 µm, 300 Å) maintained at 70 °C. Aqueous mobile phase (MPA) was composed of 0.1% TFA in milli-Q water and organic mobile phase (MPB) was 0.1% TFA in acetonitrile. Flow rate was set to 0.5 mL/min. Gradient program was set for a total run time of 8.0 min (starting at 5% B hold for 0.25 min, to 100% B at 4 min, hold at 5.8 min, back to 5% B at 5.85 min and equilibrate at 5% B till 8.0 min). Peptide was detected using UV at 220 nm.
High Resolution XRM Image Collection
The sample images were acquired using a Versa 520 XRM system (Carl Zeiss Microscopy USA). The cylindrical implant sample (Fig. 1a) was mounted vertically on a rotational stage between an X-ray source and a detector. The central portion of the implant, highlighted by a red box in Fig. 1a, was imaged to avoid potential artifact that maybe induced by mechanical cutting. To resolve the low atomic number and low-density material phases in the implant, an improved phase contrast was obtained by generating interference between the phase-shifted and unshifted X-Ray signals. The phase-shifted signals were created because of different refractive indices of different phases. The refractive index is directly proportional to the wavelength, therefore the lower the X-ray energy is, the larger the phase-shift, and therefore the better the phase contrast. While this technique required longer scanning time, it ensured sufficient contrast to differentiate API-rich region, polymer-rich region, and porosity. The X-ray source was focused on the central portion the implant rod to avoid interference from the bottom where the sample was mounted to the stage and the tilting from the tip. One X-ray radiograph was taken at the beginning using an exposure time of 0.5 s and X-ray source energy of 80 keV (c.f. Fig. 1b). Afterwards, the sample was rotated by 0.09 degrees before another radiograph was taken using the same exposure time and energy. A total of 4000 radiographs was collected iteratively in this manner. A volume of 1000 × 1000 × 1000 voxels was reconstructed using a filtered backward projection algorithm with bin averaging value of 2 with an effective voxel size of 0.5 μm. Fig. 1c shows a representative example of radial cross sections from the reconstructed volume, while Fig. 1d illustrates a 3D volume visualization of the portion of the rod imaged. All imaging studies were conducted in 3D for this project. For visualization clarity and interpretation simplicity, one 2D cross section from the 3D XRM image data will be shown on in later figures of the manuscript, unless noted otherwise.
FIB-SEM Cross Section Imaging and EDS Elemental Study
On a scanning electron microscopy system, a focused ion beam can be added to mill away some material and expose the cross-sectional features under the exterior surface of a sample. This system, referred to as focused ion beam scanning electron microscopy (FIB-SEM),
is employed in this project to confirm the structural features observed in XRM. The implant or powder samples were fixed on an SEM stub with carbon tape. To protect the particles from ion beam damage during cutting/milling, a Pt layer (~ 20 nm) was coated onto the particles using an SPI sputter-coater (Structure Probe Inc., West Chester, PA). Cross-sectioning of particles was carried out using 30 keV Ga focused ion beam (FIB) of various currents. SEM imaging was carried out using a 2 keV electron beam. Secondary electron detector (SE) and energy selective backscattered (EsB) electron detector were used for SEM imaging.
An energy dispersion x-ray spectroscopy (EDS) detector (Oxford Instrument Inc.) was to acquire X-ray spectra for elemental identification. To minimize electron–sample interaction volume (hence improving spatial resolution in elemental analysis), a 5 keV beam was used in EDS analysis, with an estimated spot size of 1 µm. The working distance for EDS was 15 mm.
AI-based Image Analytics
Digitization of all the rod samples into a large number of images was achieved by 3D XRM imaging. Image segmentation of the material phases was necessary to transform greyscale values into voxelized models for API-rich region, polymer-rich region, and porosity. Due to the low contrast between the API-rich and polymer-rich regions, conventional image segmentation methods such as greyscale thresholding did not segment the API-rich region accurately. An AI-based image processing algorithm employing supervised machine learning was applied instead. This algorithm mimics the ability of the human eyes to differentiate structural features. A feature is recognized not only based on the greyscale value of pixels, but also its relationship with its surrounding pixels defined by a basket of 10–25 mathematical filters such as Gaussian, Median, and Hessian.
The collection of the pixels reflects a unique signature of a material in response to the imaging signal, as a textural pattern based on a human expert supervision. The AI engine first learns these patterns from a human user and then uses this knowledge to automatically segment images using high performance cloud computing. After segmentation, quantitative information such as volume percentages, spatial distribution uniformity, and phase domain size distributions were computed and compared for different samples. Geometry features of the implant samples such as dimension and surface roughness were also quantified. A cloud-based image management software DigiMTM I2S
was used for image segmentation, visualization, and quantitative analysis.
Image-Based Permeability Simulation
Permeability was computed using the 3D voxel network of API-rich region and porosity reconstructed from the XRM images of the time-zero implants. After the API-rich region was numerically extracted from the imaged volume, a digital porous network, corresponding to the extracted API-rich region, was used as the input in the computational fluid dynamics simulation module in DigiMTM I2S. Fig. 2 shows a small representative 2D portion of the porous network. The simulation module solves for the Navier-Stokes equation. Each voxel in the porous network is used as a computational cell, where pressure and velocity fields are solved using a finite volume method. No flow and no-slip boundary conditions are specified on the walls between pore and polymer phase. The same wall conditions are specified on the four exterior boundaries of the cubical porous voxel network that are perpendicular to the flow direction (only two of such walls are shown in Fig. 2). Two external boundaries parallel to the flow condition are specified as pressure inlet and pressure outlet , effectively applying a pressure gradient along the flow direction. After the simulation, Darcy's law, is used to derive permeability k,
where Q is flow rate (volume per time), A is the cross-sectional surface area perpendicular to flow direction, and is the pressure gradient along the flow direction. After the permeability simulation is completed along one direction, the other two directions are simulated in a similar manner. A permeability vector is eventually calculated with three components, kx, ky, kz, corresponding to X, Y, and Z directions respectively. The magnitude of this vector, , is eventually computed and reported. This pressure driven permeability has the unit of area. The permeable phase was altered numerically to either exclude or include porosity with the API-rich region.
The permeability simulation model has been thoroughly validated in geoscience and material science applications,
An overview of all the samples investigated in this study are shown in Fig. 3. High-resolution 3D images were collected non-destructively with XRM at 0.5 μm resolution for all samples. Images of samples manufactured at the different facilities are separated by a green polyline in Fig. 3. Samples from the two facilities showed fundamentally different microstructure morphology in terms of phase dispersion, and size distribution of porosity and API-rich regions. The difference in the processing conditions between samples from the two facilities has been described in Section ``Preparation of the long acting implant via a hot melt extrusion process". These images were subjected to image analysis and microstructure quantification as shown later in Table 2. Due to the large number of samples studied, only selected samples are used to illustrate the main discussion points below. Facility B samples in general had higher porosity, more dispersed API-rich region with weaker phase contrast and less agglomeration. The polymer-rich region in facility B samples also appeared to have lower density. Thus, the contrast between the API-rich and the polymer-rich regions is lower in facility B samples under the same XRM imaging condition. Higher extrusion temperature is believed to play a major role in this different contrast morphology. At higher temperature, API and polymer are subjected to potentially stronger mixing both mechanically and inter-molecularly, leading to reduced contrast between the two. The decreased phase contrast between API-rich region and polymer-rich region is also observed when the API spray drying solution is changed from NaOH to NH4OH, suggesting that the degree of mixing also depends on the API spray drying solution.
API-rich Region Confirmation
XRM imaging identified distinctive material phases with different densities, as shown in Fig. 4a. Fig. 4b shows a magnified image of a small region from Fig. 4a, where the three separate material phases are seen inside the implant with distinctive greyscale contrasts: black, dark grey, and light grey. The black contrast phase corresponds to the air void, as it has the lowest possible density. As X-Ray attenuation is predominantly dictated by the density and elemental atomic number of the material, brighter greyscale values correspond to material phases with a higher true density. The API has a higher density (1.34–1.36g/cc) than the PLGA polymer (1.29–1.3g/cc). It is therefore reasonable to assign the brighter phase in the XRM images as API-rich region. Thus, the dark grey and the light grey regions of the image were assigned to the polymer-rich and API-rich respectively. To further confirm the phase assignments, in Fig. 4a, rod sample D520 was co-imaged with the corresponding API powder sample. The pure API particles outside the implant sample (labeled with red letters “O”), have a greyscale contrast brighter than the predominant dark grey phase inside the implant. By derivation, the phase inside the implant that is brighter should be the API or API-rich region.
One complication with the XRM implant-powder co-imaging experiment is the beam hardening artifact on the exterior surface of the implant, marked by a yellow arrow in Fig. 4b. There is a substantial density discontinuity between the outside air and the implant sample. Furthermore, the different material phases inside the implant have relatively small density difference, requiring lower energy and longer exposure time. This imaging configuration yields a bright “shell” around the implant outer layer, due to X-Ray beam hardening. It introduces an uncertainty on whether the particles labeled “O” were indeed dense, or negatively impacted by beam hardening on the implant sample surface.
In order to definitively confirm the assignment of the material phases, a FIB-SEM experiment was conducted using the system illustrated in Fig. 4c. After FIB milling, the subsurface cross section was exposed, where an SEM image and an EDS study were conducted. Based on this FIB-SEM study the identity of the three material phases are confirmed: porosity (black phase), polymer-rich (dark grey phase), and API-rich (light grey phase). Note that the porosity phase in the FIB-SEM image is impacted by electron charge accumulation, which leads to the brightest contrast. It does not alter the confirmation of porosity phase and has no impact on the confirmation of API-rich and polymer-rich regions.
Elemental weight % averaged over 15 Spot EDS measurements in Table 3 confirms the phase contrast between polymer and API phases through the strong signals of sulfur, and the presence of nitrogen signal that are unique to the API domain. In comparison, the polymer domain has no nitrogen signal, and orders of magnitude weaker sulfur signal. Some sulfur signal still shows up on polymer domain due to some degree of polymer-API mixing at a scale smaller than the EDS spot size. Such mixing cannot be captured with nitrogen signal because it is already pretty weak in the API domain.
Table 3EDS of chemical composition weight % measured from Fig. 4d. Fifteen measurements for API phase fifteen spots similar to the features marked by the letters “A”, are averaged. Similarly, for polymer phase fifteen spots similar to the features marked by the letters “P”, are averaged.
The XRM images of two implant samples, D520 and D624, the XRM images of the corresponding powder APIs used in making these two implants (API-NaOH and API-NH4OH), and the corresponding SEM images of the powder APIs, are shown in Fig. 5. XRM images of all four samples were scanned at 0.5 µm resolution. SEM images were collected at 0.1 µm resolution (2000x magnification).
While phase contrast between the API-rich and the PLGA polymer-rich regions is clearly observed in both implants (Fig. 5a andd), the degree of phase contrast appears to be higher for the D520 sample. D520 sample showed API-rich regions that were well-dispersed inside the implant with minimal agglomeration while D624 sample showed a higher degree of API agglomeration. API-NaOH (Fig. 5b) had dominantly spherical particles with internal voids
and a lower degree of agglomeration (arrow), while API-NH4OH (Fig. 5e) showed dominantly solid particles with no voids visible at the current resolution, and a higher degree of agglomeration. The agglomeration patterns observed in the loose API powder sample corresponded well with the agglomeration patterns in the implant sample, i.e., a larger agglomeration in the API-NH4OH in the implant corresponded with larger degree of agglomeration of the API-NH4OH particles. SEM images (Fig. 5c andf) shows similar observation, though the agglomeration of API-NH4OH (Fig. 5f) is less statistically abundant due to the significantly smaller field of view (hence much less number of particles investigated). It is important to note while D624 clearly shows higher degree of agglomeration than D520, suggesting that the counterion difference maybe an explanation, the higher drug loading of D624 can also contribute to the higher degree of agglomeration. It is also noted that the stronger agglomeration is only apparent in high drug loading API-NH4OH samples from facility A. Two facility B samples that have the same 25% drug loading but used API-NaOH and API-NH4OH, i.e., sample D729 (Fig. 3k) and sample D824 (Fig. 3m), do not have obvious agglomeration.
Phase Segmentation with AI
After the identification and confirmation of different material phases, an artificial intelligence (AI)-based image segmentation method is used to segment the phases into 3D labels, from which additional phase quantification and image-based simulations can be conducted. An AI-based approach is critical to obtain acceptable phase segmentation results for low density and low contrast material such as the implant samples in this project. Fig. 6a shows one 2D cross section image of the XRM scan, using sample D520 as an example. The API-rich region is indicated by “A”, polymer phase is indicated by “P”, and porosity phase is indicated by the blue arrows. Through the AI segmentation method, porosity (blue), API-rich (red), and polymer-rich (grey) regions are segmented with one computing session (Fig. 6b). Conventional threshold segmentation methods, in comparison, did not successfully label these phases. Fig. 6c shows a greyscale intensity threshold value of 26,500, where the API-rich region is under-segmented. When the threshold value is decreased to 26,100, in order to include more pixels into API-rich region segmentation, it is clear that the API-rich region is over-segmented (Fig. 6d). White circles in Fig. 6(b), (c) and (d) are used to highlight an example segmentation of a API-rich region. AI segmentation out-performs conventional thresholding method in both accuracy and reliability (Fig. 6b). For a 16-bit unsigned data type, the maximum intensity value is 215-1=65,535. The intensity variation of 26,500-26,100=400, is only 400/65.535 = 0.6% of the total intensity range. This suggests that the accurate segmentation is required with very high accuracy within a very small intensity range, where finding the right threshold not possible
. AI segmentation on the full 3D dataset does require additional computing resources, which is readily managed by cloud computing.
Phase Domain Size Distribution and Content uniformity
Volume % of API-rich regions
Attributes relating to the microstructure of the implants, such as volume percentages of API-rich region and porosity, are summarized in Table 4, and the representative images from the corresponding samples are shown in Fig. 3.
Table 4Imaging derived properties of samples investigated in this study. The green line and grey shading differentiate samples manufactured at different facilities.
The volume %s of API-rich region in the rods calculated from the XRM images were compared to the weight-based drug loads that was specified by the formulation. The correlation between volume and weight %s is expected to be linear for molecularly phase-separated materials. However, there are a few uncertainties that can limit the accuracy of this correlation. XRM imaging, in order to capture the full diameter of the implant, has a limited resolution of 0.5 µm. In typical XRM characterization, 3–5 voxels are required to resolve a feature. Hence, XRM can only resolve features larger than approximately 2 µm. For microstructure domains smaller than 2 µm, they are considered as sub-resolution features. As each imaging voxel can only be assigned to one material phase, when that voxel was primarily occupied by the API, it was assigned to the API-rich region even though a small amount of polymer phase and porosity might be present inside the 2 µm feature. Earlier Fig. 4d showed a yellow contour with an agglomeration of several smaller API particles, as an example of such sub-resolution API-rich feature. Consequently, overcounting of the volume percentages is expected. As an additional limit on the accuracy of volume-weight conversion, the densities of the polymer phase might be altered based
on processing conditions for the implant. Despite of these uncertainties, the volume % of the API-rich phase determined using XRM was linearly correlated to the weight % as expected (Fig. 7). This data indicates that image segmentation at the 0.5 µm resolution is sufficiently accurate for calculating implant attributes.
To ensure product quality and consistent in vivo performance, it is desirable for the API-rich region to be evenly distributed inside the implant. While such an assessment would be extremely difficult with traditional characterization approaches, it is readily accomplished with 3D non-invasive micro-imaging analytics. Qualitatively, the distribution of the API-rich region in the implant was assessed visually to confirm that the API-rich region did not have obvious localization. From Fig. 3, the majority of the samples studied showed reasonably good API uniformity at a qualitative level, except for two higher drug loading API-NH4OH samples from facility A. The uniformity was further quantitatively assessed along both radial and axial directions on the segmented API-rich region, as shown in Fig. 8. Fig. 8a and b shows axial and radial distribution of API-rich regions, while 8c illustrates the schematics of how the calculations were conducted. Three samples, D731, D730 and D729, were selected from facility B to demonstrate the quantitative uniformity analysis. These samples from facility B are also manufactured with better quality control. They are closer to the final product, hence bear more significance in uniformity control. One sample, D624, was selected from facility A as a comparison. In Fig. 8a, along axial direction, the volume percentages of the API-rich region were computed in each cross section. Good uniformity for the three samples from facility B were observed at different drug loads. Sample D624, in comparison, has a higher concentration of API-rich region on one side of the sample. The volume % of the API-rich region varies substantially from 43 to 62%. For the radial direction, a cylindrical coordinate system was adopted, where a radial cylindrical cut was digitally computed. The cylindrical cut was then unwrapped into a 2D image where the API volume % was computed. Fig. 8b, similarly shows good radial drug uniformity for samples D729, D730 and D731 from facility B. Larger fluctuations in volume %s toward the outer surface (radius greater than 220 µm) was observed for all samples. This is due to the deviation of the geometry of the samples from a perfect cylinder, and the variations of the sample radius. In comparison, sample D624 has higher API-rich region of 60% in volume % near the center of the rod. Toward the outer region, the API-rich volume fracture decreases substantially, with the volume % as low as 47% toward the outer surface of the rod. The uniformity difference between facility A and B is fairly pronounced.
API-rich region domain size distribution (SD) inside implant
Domain size of the API-rich region can affect the distribution of the API within the polymer matrix and its release performance. It is therefore critical to understand the domain size distribution (SD) of API-rich region inside the implant. While the SD of API-rich region inside the implant are difficult to be reliably determined with traditional approaches, it can be readily characterized with 3D XRM imaging The average value of D10, D50, D90, and porosity with corresponding statistics for implants made from API-NaOH and API-NH4OH and corresponding values for the as-received API are shown in Table 5. The following inferences can be drawn by comparing the size and porosity values of the implants made with the two different APIs: (i) The coefficient of variations (CV) for D10, D50, D90, and % porosity remain fairly consistent for the two API types with CV percentage being the highest for % porosity, (ii) Implants made from API-NaOH showed smaller SD than the implants made from API-NH4OH, and (iii) implants made from API-NaOH had higher % porosity than the implants made from API-NH4OH. The SD values of the API-NaOH and API-NH4OH powders as measured by laser diffraction were similar (Table 5). This indicates that API-NaOH has a lesser propensity to aggregate and form API-rich region when compared to API-NH4OH. This was confirmed by the XRM images of the loose API powder samples (Fig. 5b and e). Thus, it is clear that the counter ion in the formulation has a profound impact on API particle characteristics in the implant, and consequentially the implant performance.
Table 5Particle size distribution and porosity statistics based on API source used in the implants.
Note due to the resolution limitation of the XRM, primary API particles cannot be resolved. The size distribution computed from the XRM images are clusters of APIs and API aggregates. The primary API particle characterization inside the implant is the subject of future FIB-SEM report. Despite this limitation, the size and interconnectivity of the API domain, particularly compared across different samples, remain to be significant.
The desired dimensions of these cylindrical implants are 500 µm in diameter. Due to the small diameter, its impact on final drug product performance is significant as a small variation in diameter could lead to a substantial change in surface area, and thus the effective area for API release from the rod. Therefore, it was deemed critical to control the diameter of the rod to between 470 and 530 µm throughout the rod. Surface imperfections should be minimized and/or prevented. Two samples, D520 and D824 were selected to compare the geometrical qualities of the samples manufactured at the two facilities. Fig. 9a shows a cross section of the 3D XRM image corresponding to sample D520, manufactured at facility A. Fig. 9b shows the same for sample D824, manufactured at facility B, where better quality control was implemented. For each radial cross-section, a length (longest Feret diameter) and a width (shortest Feret diameter) were measured, and an equivalent circular diameter was estimated, as shown in Fig. 9c. Sample D520 has diameter variations between 392 and 431 μm with an average diameter of 419 μm, substantially smaller than the required diameter, while sample D824 has diameter variations between 487 and 518 μm with an average diameter being 497 μm., satisfying the diameter requirement. The distribution of the rod perimeter for each cross-section was also computed in Fig. 9d. Similarly, larger range of variation is observed in Sample D520. Additional geometric parameters such as circularity and elongation can also be derived from these measurements.
The smoothness of the rod surface, closely related to the surface area, can also be assessed through the imaging measurements. Rugosity is a measure of small-scale variations of amplitude in the height of a surface or surface roughness. For each cross section, along the outer perimeter, rugosity was computed as , where is the actual surface area, and is the geometrical surface area that is calculated by smoothing the perimeter to remove all small local fluctuations. For a perfect cylinder, rugosity is 0. Sample D824, manufactured at facility B with better quality control, has an average rugosity of 0.22. In comparison, sample D520 has both larger rugosity fluctuation, and larger average rugosity value of 0.28. Unparalleled geometric characterization can be achieved with XRM imaging data. This in turn can be used to optimize manufacturing process to achieve the desired quality attributes and ensure reproducible in vivo performance of the implants.
Impact of Formulation and Process Parameters on Implant Microstructure
Changes in formulation composition and/or process conditions could lead to variations in the implant microstructure and can thereby affect the desired quality attributes of the implants. As indicated in Table 1, implants described in this study were formulated using two different PLGA polymers (85/15 vs. 75/25, both with 100 kDa molecular weight) and API with two different counter ions (Na+ and NH4+). The implants were extruded at two different temperatures (90 °C vs. 120 °C) at various drug load ranging from 15 to 35%. Fig. 10 show the interplay among these factors in terms of how they affect implant microstructure (porosity).
As discussed in Section ``API-rich region confirmation", the two different sources of API have a significant impact on domain size distribution of API-rich region and porosity. For each API source shown in Fig. 10a, implants manufactured with higher extrusion temperature (120 °C, circular symbols) had higher porosity than the implants manufactured with lower extrusion temperature (90 °C, square symbols). For both API sources extruded at 120 °C, increasing drug load also led to higher porosity. Overall, the average porosity of the samples made with API-NaOH was higher than that of the samples made with API-NH4OH. While the exact origin of this porosity difference is a subject of continued investigation, it is likely related to the difference of the volatility of the counterion post spray drying. The Na+ counterion tends to remain in the API post spray drying while NH4+ has a tendency to leave the API due to its volatility. The presence or absence of the counterion can influence the microstructure in the API and consequently of the API in the implant.
For each API source, Fig. 10b shows the impact of L/G on the implant porosity. For a given API source and drug load, implants made from PLGA co-polymer with a L/G ratio of 75/25 had higher porosity than those made with a L/G ratio of 85/15.
Image-Based Transport Property Modeling
The interconnectivity of the API network in implants typically controls the release of API in a monolithic dispersion-controlled release system.
If polymer is neither swollen or degradable, API network interconnectivity and porosity are the primary factors that affect release throughout the entire release duration. While the use of degradable polymers such as PLGA complicates the release behavior, still the initial drug release from such systems is dominated by the API network.
The XRM imaging data can be used to characterize the interconnectivity of the API-rich region of this API network, which in turn can be used in the simulation of system permeability. The permeability values of the implants were computed with and without the consideration of porosity. The permeability with and without porosity consideration as a function of overall drug load is shown in Fig. 11a and b respectively. When porosity of the implant was not considered, higher drug load led to increased network interconnectivity and hence a higher permeability, as expected. At 25% drug loading, where data was available from both API sources, the implant made from API-NaOH showed higher network interconnectivity efficiency and permeability when compared to the implants made from API-NH4OH. The effect of extrusion temperature on permeability without porosity was negligible at 25% drug loading.
When the porosity of the implant was considered, it clearly affected the interconnectivity and corresponding permeability. Increased porosity provides additional pathways for network interconnectivity when compared to just interconnected API domains. Thus, increased porosity leads to increased sample permeability and thereby leading to higher API release rate. Data in Table 4 shows that the implants have a range of porosity values from 0.2 to 11.46%, with an average of 1.4% for facility A samples, and an average of 6.7% for facility B samples. While the porosity difference between implants is seemingly low, it leads to a substantial increase in permeability, from 1.5% increase in a low porosity implant to 99.1% increase in a high porosity implant. Furthermore, the spatial arrangement of the porosity has an important role to play. Sample D729 has 8.56% of porosity, slightly more than sample D731’s 7.2%. However, sample D731’s permeability is increased by 98.2% from the porosity, significantly more than sample D729’s 47.5%. This is likely due to differences in porosity spatial distribution and its relative arrangement with different drug loading in these two samples. Fig. 11b demonstrates a roughly linear dependence between porosity increase and permeability increase. This porosity impact on permeability was more pronounced when porosity was at least 3% or higher, which was dominantly associated with the API-NaOH implants.
It is also noted that from Table 2, the manufacturing conditions of sample D520, D621, and D622, such as facility, polymer L/G ratio, drug loading, extrusion temperature, and API counter ion are all the same. However, in Table 4, the sample D520 has a higher porosity of 3.4% compared to the D621 (0.2%) and D622 (0.4%). Sample D520 was manufactured from a batch of API that had a higher sodium content (3.0%) while samples D621 and D622 were manufactured from API that had a lower sodium content (1.7–2.0%). It is likely that the sodium content difference in the APIs used to make these rods caused the change in the porosity.
Determining quality attributes of LA implant formulations such as API-rich region domain size distribution, axial and radial distribution of the API-rich region, porosity, and pore size distribution, as well as geometric variations, are difficult if not impossible to do with traditional characterization techniques. We used XRM imaging on as-received drug product samples to qualitatively characterize implant geometry, drug domain distribution, and microporosity. AI-based image segmentation reconstructed the microstructures into numerical models from which quantitative data on geometric variations, drug dispersion uniformity, and API aggregate size was obtained. The quantitative information from images can be used to adjust the process and/or formulation to provide implants of acceptable quality. For example, subtle differences in the API such as the counter ion show marked differences in porosity and the size distribution of API-rich region in the implant. Moreover, geometric variations in the implant samples can be characterized in great detail using the image-based approach, which again cannot be achieved with traditional imaging techniques like light microscopy. Cross sectional variations of Feret length and width, and surface roughness, can be easily determined from the 3D reconstructed images. Finally, important microstructure attributes such as phase behavior, porosity, and pore size distribution can be easily determined using the segmented images. Ultimately the implant microstructure was found to be dictated by the complex interplay of API counter ion, polymer type, processing temperature, and drug loading. Since this information can be determined in a non-destructive manner on the as-received implant sample, the XRM based approach can be utilized to gain a fundamental mechanistic understanding between formulation/process parameters and drug product performance.
As LA formulations are intended to release drug over long periods, developing in vitro release testing methods that mimic duration of release in appropriate media can result in long development times. In order to circumvent this issue, accelerated in vitro release methods are typically developed. However, accelerated release methods typically do not capture the underlying physics of drug release from the implant and merely serve as correlational tools. Drug release from the implant is controlled in part by the API domain interconnectivity which in turn depends on drug loading, API domain size, and porosity. These parameters can be readily quantified from the segmented XRM images. Subsequently, the efficiency of the API network interconnectivity can be characterized by permeability. In cases of a non-degradable matrices where API and polymer are phase separated, the permeability values can be used directly to predict the API release profile from the implant. However, in cases where the matrix is degradable, as in our current situation, the polymer degradation must be combined with permeability values in order to predict API release. Thus, in theory, XRM based methods can also be used to predict API release from the implant leading to shortened development cycles.
While advantages of using XRM to expedite the development of LA formulations were discussed in the previous sections, it is also important to understand some limitations associated with the technique. (i) For polymeric samples with relatively low glass transition temperatures (Tg) such as PLGA, care needs to be taken to reduce the potential X-ray induced damage to the sample. On the other hand, the lower X-ray dosage often required for these types of materials poses challenges in achieving suitable contrast and signal to noise ratio, necessary to differentiate the material phases with small density differences and low atomic numbers. (ii) Due to the resolution limit, phases smaller than 1.5–2 μm cannot be resolved by XRM. For example, due to the small sizes of starting API particles (See Table 5), the features of the as-received API powders are unlikely to be resolved with XRM. Rather, only API-rich regions can be characterized with XRM. Furthermore, although XRM has been successfully demonstrated to differentiate amorphous API from crystalline API in certain drug systems,
the API domain in this project is too small to support any reliable crystallization investigation. In order to overcome this resolution limit with XRM, other 3D imaging techniques with higher resolution such as FIB-SEM
have been be applied in pharmaceutical drug material characterization. The resolution of FIB-SEM can be as high as ~3 nm, revealing microstructures that are not resolvable under XRM. The tradeoff is that the sampling volume will be much smaller. The combined application of both XRM and FIB-SEM imaging technologies is expected to be a powerful tool that can further aid in the development of LA formulations.
Fig. 12 shows a generalized workflow for characterizing implant attributes and for predicting release profiles from implants using imaging techniques. The combination of high-resolution imaging and image analytics can be used to measure key implant attributes that were hitherto unmeasurable using conventional techniques. This unparalleled ability to measure key implant attributes with XRM is expected to greatly improve and accelerate the development of LA implants.
Finally, image-based drug product characterization produces a large amount of data, including raw images at various resolutions, derived images after image processing and segmentation, analysis results, and simulation. The associated workflow can involve many steps, which renders efficient management and tracking of workflow, user's activity, and computational resources very important.
In this paper the utility of XRM-based microstructure characterization for implants was demonstrated. Several monolithic implants prepared using hot melt extrusion process were analyzed by XRM. XRM images were segmented into different material phases using an AI-based approach. A full 3D tomographic reconstruction of segmented 2D cross sections was used to derive quantitative information about implant attributes.
The use of advanced XRM image analysis was shown to be very useful to gain a mechanistic understanding of how process and formulation parameters impact implant attributes, and to shorten development timelines associated with LA formulations. Moreover, XRM is a non-destructive method that allows for samples to be assessed without contact with external environment
and to be reused for other purposes as needed. In comparison with traditional physicochemical characterization approaches, image-based characterization is an attractive alternative in characterizing long acting dosage forms, with considerably improved cost and time efficiency.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
The dexamethasone drug delivery system: indications and evidence.