Characterization of pore structure of polymer blended films used for controlled drug release

Journal of Controlled Release 222 (2016) 151–158

Contents lists available at ScienceDirect

Journal of Controlled Release journal homepage: www.elsevier.com/locate/jconrel

Characterization of pore structure of polymer blended films used for controlled drug release Henrike Häbel a,f,⁎, Helene Andersson b,c,f, Anna Olsson d,f, Eva Olsson d,f, Anette Larsson e,f, Aila Särkkä a,f a

Chalmers University of Technology and University of Gothenburg, Department of Mathematical Sciences, SE-412 96 Gothenburg, Sweden SP Food and Bioscience, Structure and Material Design, Box 5401, SE-402 29 Gothenburg, Sweden Chalmers University of Technology, Department of Materials and Manufacturing Technology, SE-412 96 Gothenburg, Sweden d Chalmers University of Technology, Department of Applied Physics, SE-412 96 Gothenburg, Sweden e Chalmers University of Technology, Department of Chemistry and Chemical Engineering, SE-412 96 Gothenburg, Sweden f SuMo BIOMATERIALS, VINN Excellence Center, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden b c

a r t i c l e

i n f o

Article history: Received 25 August 2015 Received in revised form 2 December 2015 Accepted 8 December 2015 Available online 11 December 2015 Keywords: Pore shape Permeability Image processing Porous material Scanning electron microscopy

a b s t r a c t The characterization of the pore structure in pharmaceutical coatings is crucial for understanding and controlling mass transport properties and function in controlled drug release. Since the drug release rate can be associated with the film permeability, the effect of the pore structure on the permeability is important to study. In this paper, a new approach for characterizing the pore structure in polymer blended films was developed based on an image processing procedure for given two-dimensional scanning electron microscopy images of film crosssections. The focus was on different measures for characterizing the complexity of the shape of a pore. The pore characterization developed was applied to ethyl cellulose (EC) and hydroxypropyl cellulose (HPC) blended films, often used as pharmaceutical coatings, where HPC acts as the pore former. It was studied how two different HPC viscosity grades influence the pore structure and, hence, mass transport through the respective films. The film with higher HPC viscosity grade had been observed to be more permeable than the other in a previous study; however, experiments had failed to show a difference between their pore structures. By instead characterizing the pore structures using tools from image analysis, statistically significant differences in pore area fraction and pore shape were identified. More specifically, it was found that the more permeable film with higher HPC viscosity grade seemed to have more extended and complex pore shapes than the film with lower HPC viscosity grade. This result indicates a greater degree of connectivity in the film with higher permeability and statistically confirms hypotheses on permeability from related experimental studies. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Porous polymer blended films are often used as pharmaceutical coatings since they can provide a wide range of structures with different properties favorable for controlled drug release [1]. In order to understand and control mass transport properties like permeability, it is essential to characterize the pore structure within such films. While methods to experimentally study the porosity of a material have long been available [2], today's imaging techniques such as scanning electron microscopy (SEM) open up new opportunities to characterize the pore structure with more attention to details using tools from image analysis including, for instance, binarization and pore boundary detection [3,4]. These tools can be used to study porosity and pore shape. The pore shape, in turn, can be related to pore tortuousity and connectivity, which have previously been identified as important factors affecting

⁎ Corresponding author. E-mail address: [email protected] (H. Häbel).

http://dx.doi.org/10.1016/j.jconrel.2015.12.011 0168-3659/© 2015 Elsevier B.V. All rights reserved.

mass transport and overall releasability of a drug [5]. Therefore, the development of appropriate image processing procedures to extract the pore structure and perform statistical image analysis of its detailed characteristics have become of large interest. In this article, we statistically compare pore characteristics of blended films of two of the most common cellulosic polymers used in controlled release formulations, namely ethyl cellulose (EC) and hydroxypropyl cellulose (HPC). Such bio-based films are non-toxic, non-allergenic and have good film forming properties and stability [6]. Whereas EC is water insoluble, HPC is generally soluble in water or in the gastrointestinal tract at room temperature (0-solvent at about 41 °C) and can be used as a pore former [7]. The two polymers are dissolved in a common solvent, which evaporates during film spraying resulting in phase separation. In this way the film structure forms and the pores result from subsequent HPC leaching [8,9]. Hence, the HPC-rich domains serve as a template for pores and determine their size and shape. There are several factors influencing the formation of the pore structure of EC/HPC films such as film processing parameters [8], polymer blend composition [6] and polymer viscosity grade [10].

152

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

In this study, a fixed polymer blend composition with 30% w/w HPC and two different HPC viscosity grades were used to produce two types of films. The polymer blend ratio was chosen to ensure that a percolating pore system forms, where the main release mechanism is diffusion through the pores. The choice was based on results from previously studied systems of EC and HPC [6] indicating that 30% w/w exceeds the critical HPC concentration for obtaining a connected, percolating pore network with channels going from one film side to the other. For films with HPC concentrations below the percolation threshold, micro-structural characteristics may be of less importance for drug release due to the possibility of a convective release process occurring through cracks in the film. For the two films studied here, experiments on leakage of HPC indicated no prominent differences between the films, whereas a great differences in permeability was measured. The leaching experiments showed an expected high release of almost all HPC from both films confirming that the percolation threshold had been exceeded and that the porosities were similar. Hence, the results suggested that the pore shape is responsible for differences in permeability for the two films. However, mere visual inspection of microscopy images showed similar pore structures. That is why it was of great use to extract and quantify the pore structure and identify suitable pore characteristics in order to explain the difference in measured permeability. The aim of this article is to present an image processing procedure to extract and analyze characteristics of porous materials. The pore characterization is used to draw general conclusions about the pore structure in EC/HPC films and its effect on permeability. Given two percolating pore systems with channels allowing for drug release by diffusion, we focus on the shape of a pore extending previous research on the pore size distribution [3]. As a result, our study has considerable potential to statistically prove what experiments failed to show, namely a difference between the pore structures of the two compared films.

fluidized-bed chamber at AstraZeneca R&D Mölndal, Sweden, following the procedures as described by Larsson et al. [11]. Before spraying, the center part of the Teflon drum was covered with a strip of plastic tape (Deer Brand, Four Pillars) to facilitate the film removal. After a drying time of about 50 min, the films were removed from the drum. As a consequence of the preparation, the films are relatively homogeneous in each sprayed layer, but inhomogeneous in the drum-to-air side direction showing periodic layers of different pore structures [8]. 2.2. Water permeability In this study, the water permeability P of EC/HPC films was used as a measure of mass transport through their pore structure. In particular, the net transport of water from regions of relatively high concentrations c1 to regions with relatively low concentration c2 by random molecular motion in a thin film of thickness h was considered, such that

P¼

h ∂m : Aðc1  c2 Þ ∂t

Here, A refers to the surface area of the film where mass transport denotes the mass transfer rate along the cross-section occurs and ∂m ∂t direction of the film [12]. Water permeability through the EC/HPC films was measured using radioactively labeled molecules, tritiumlabeled water (Perkin Elmer Inc., USA) as the diffusant according to the procedure presented by Andersson et al. [10]. As Table 1 shows, the L film with higher molecular weight was about three times more permeable than the SL film even though the weight reduction after HPC leaching was almost the same for both films.

2. Material and methods

2.3. Data

2.1. Preparation of films

The data analyzed in this study were two-dimensional field emission scanning electron microscopy (Leo Ultra 55 FEG SEM by LeoElectron Microscopy Ltd., Cambridge, UK) images of film cross-sections. Four about 2 mm wide and 1 cm long pieces from the central part of each film were extracted. The film pieces were embedded in epoxy-glue (Epoxy Rapid, Bostik) in 1.5 cm × 0.5 cm × 0.5 cm molds and the resulting stems of glue were cut in an ultramicrotome (Powertome XL, RMC products, Boeckeler Instruments Inc., Tucson, Arizona) to expose the film crosssections. It should be noted that a smoothening effect of the exposed film surface morphology has been observed to occur as a result of the ultramicrotome cutting. The area fraction of the exposed surface corresponding to pores appeared to be reduced as a consequence of the cutting. However, the cutting was necessary to obtain flat surfaces suitable for the quantitative image analysis. The newly cut and smooth film cross-sections were exposed to water to allow HPC leaching and to reveal the pore structure in the films. The leaching procedure was done by mounting the glue stems, cross-sections side down, in a beaker with water (approximately 300 ml per film piece) under rigorous stirring for at least 24 h. The stems were attached to aluminum pin stubs using Pelco ® Colloidal Silver Liquid (TED PELLA INC.) with the exposed film cross-section surface facing up. Finally, the stub was coated with a thin layer of gold and examined in the SEM. From each of these cut film surfaces, three strips of the cross-section sufficiently far away from the edge of the film and at least 300 μm apart from each other were selected for microscopy and further analysis. All together 12 samples (image strips) from the SL and L films were taken. The pore structure from each of these 2 ⋅ 4 ⋅ 3 = 24 strips was imaged by taking overlapping SEM gray-scale images of size 1024 by 691 pixels (ca. 37.9 × 25.6 μm) from where the film initially touched the Teflon drum (drum side) to the other side (air side). Fig. 1 gives an overview of the data setup.

The two films investigated in this study were prepared by dissolving two cellulose derivatives in hydrous ethanol (95% v/v, Kemetyl AB, Sweden) at room temperature under stirring overnight. These polymer blended solutions contained 70% (w/w, dry basis) ethyl cellulose, EC, of viscosity grade 10 cps (29 103 g/mol, Dow Wolff Cellulosics GmbH, Germany) and 30% (w/w, dry basis) hydroxypropyl cellulose, HPC. HPC in two different viscosity grades, HPC-SL and HPC-L, were kindly provided by Nisso HPC, Nippon Soda Co. Ltd., Japan. The molecular weight Mw for the EC and HPC viscosity grades were determined by size exclusion chromatography as described by Andersson et al. [10]. Henceforth, the film with lower HPC molecular weight is referred to as SL film and the film with higher molecular weight as L film adopting the naming of pharmaceutical (viscosity) grades of Nisso's HPC. Details for both films are given in Table 1. The polymer concentrations of the solutions were chosen to be 6.5% w/w (EC/HPC-L) and 7.7% w/w (EC/HPCSL) in order to have equal average viscosities of the solutions, which facilitates the use of similar processing parameters during spraying [10]. The blended polymer solutions were sprayed onto a rotating cylindrical Teflon drum with a moving atomizer nozzle in a modified

Table 1 Weight average HPC molecular weight Mw in 103 g/mol, average weight reduction in % w/ w after HPC leaching and average water permeability in 10−10 m2/s. Standard deviation is given within parenthesis. All averages are of three replicates, except the water permeability of the SL-film, which is of two replicates. Film

HPC Mw

Weight reduction

Permeability

SL L

55 (0.65) 83 (0.77)

28.12 (0.13) 28.92 (0.17)

0.42 (0.01) 1.30 (0.03)

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

Fig. 1. Cross-section of EC/HPC film with magnified central part of sample strip covered by overlapping SEM images from drum-to-air side.

2.4. Image processing In order to characterize the pore structure of the SL film and L film samples, the solid (EC-rich) and pore (former HPC-rich) phases were classified using tools from image analysis in three steps. Firstly, for a comprehensive understanding of the pore structure as a function of film depth from drum-to-air side, it was crucial to merge the images covering a whole sample strip. Secondly, a suitable gray-level threshold separating the solid and pore phase was determined for the segmentation of the images. In a third step, the segmented binary images were further processed to extract the pore characteristics of interest. All steps were conducted in MatLab R2011b and R2013b essentially using the Image Processing Toolbox. A detailed description of the applied MatLab functions can be found in the MathWorks® documentation [13]. For the first step, there are fast standard methods available for merging two or more images. Here, two overlapping SEM images were consecutively merged together starting from the drum side. For this purpose, the second image was transformed according to a conversion matrix H such that for the location vectors of key-points pi from the first and pj from the second image, it holds pi ¼ H i; j p j : The matrix H was estimated by using the random sample consensus (RANSAC) algorithm [14]. The key-points were located at corners detected by Harris' method in the first and, respectively, second considered image as depicted in Fig. 2. The change of the average intensity in a test window caused by small shifts in various directions was calculated and Harris corners were detected at locations with a significant intensity change in all directions [15]. Feature vectors were assigned to each key-point according to the Scale Invariant Feature Transform method [16]. Subsequently, key-points from two images were matched when the Euclidean distance of their feature vectors was below a certain threshold [17].

153

Since the SEM images were not perfectly aligned, only a central part of height 771 pixels (ca. 28.6 μm) contained in all images was used in the analysis. Due to the homogeneity of the material in each film layer, there was no loss of information caused by this cropping. Subsections of size 315 by 771 pixels (ca. 11.7 × 28.6 μm2) were extracted after merging such that a sequence of non-overlapping film subsections covering the whole film strip from drum-to-air side was created. The images of the outmost film layers were manually cropped in order to obtain rectangular film subsections only containing pore structure. Consequently, the first and last subsection can vary in size. Prior to the segmentation of the solid and pore phase in the second step, each subsection was filtered. Filtering means that the image in its matrix representation I = (Ii,j) is transformed using matrix operations. Filtering methods often depend on a specified neighborhood around each pixel, where a n × m neighborhood corresponds to a n × m matrix. Following the suggestions by Schuster et al. [18], we developed an image processing procedure without edge detection since some pore contours were smeared over during ultramicrotome cutting. Furthermore, it was important to classify the whole interior of a pore at the surface level in order to capture information only at the current surface and not of the pore structure in lower layers inside the sample visible in some pores. The procedure consists of several steps. First, pixelwise adaptive Wiener filtering with a 5 × 5 neighborhood was used for Gaussian white noise reduction. Secondly, the contrast was increased by top hat filtering using a pixelized disk with a diameter of 5 pixels to create a morphologically open image, which was added and again subtracted from the original image. Thirdly, columnwise neighborhood minima were calculated to amplify dark structures based on sliding 3 × 3 neighborhoods. For the actual segmentation, the filtered subsections were divided into 9 parts, where each part was binarized. In a binarization, pixel values below a certain threshold are set to black and above it to white, respectively. For the given data, the threshold was chosen to be g = gm ⋅ ga, where gm was manually chosen according to the expected pore area fraction and ga automatically by using Otsu's method. Otsu's method finds the threshold which minimizes the variance of observed intensities below and above it in a gray-level histogram [19]. In the third and last image processing step, the binary images were smoothened by applying a median filter with a 2 × 2 neighborhood, where the intensity at each pixel is set to the median of the intensities of itself joint with the neighborhood around it. Finally, the images were inverted such that the pores were considered as white foreground objects in the image analysis, whereas the solid phase was set as black background. Furthermore, objects smaller than 40 pixels (ca. 55 nm 2) were declared artifacts and removed. Consequently, some small pores were falsely removed, which should not cause any problems in the subsequent analysis due to their relatively small affect on permeability. For further analysis, skeleton images were produced reducing each object to a minimally connected line segment preserving the number of objects and holes in each image [20]. Fig. 3 gives an example of a

Fig. 2. Schematic Harris corner detection and key-point matching: the small squares represent test windows comparing the change of average intensity in various directions. At locations with a change in all directions, a Harris corner has been detected as shown for two examples of matched key-points placed at detected Harris corners.

154

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

Fig. 3. Left: magnified part of the original SEM image, right: binary image with a skeleton (gray) as well as two branching points (crosses) and four endpoints (circles) inside the largest depicted pore.

magnified small section of one sample and the final binary image. In addition, the skeleton is depicted, which consists of lines connecting branching points and endpoints. 2.5. Pore characterization In order to characterize the pore structure as a function of film depth, each film subsection was divided into five equally sized parts with the exception of the outmost subsections. Due to the manual cropping of the images of the film ends, the first and last part varied in size. In total, there were 12 to 13 original SEM images per sample resulting in 60 and 65 parts, respectively. Pore characteristics CðtÞ were evaluated as a function of depth t = i ⋅ 2.3μm for part i = 0 , … , n - 1 , n ∈ {60, 65} starting at the drum side. The pore structure was characterized according to two aspects. First, we studied common characteristics such as pore area fraction, pore size and number of pores. Secondly, we scrutinized pore shape characteristics describing their orientation in the film and their shape. In the following, the considered pore characteristics are listed. • Pore size characteristics - Pore area fraction: number of white pixels divided by the size of the image in pixels - Pore size: number of pixels within an object; this number was converted to μm2 by dividing by the squared resolution, which was 27 pixels per μm - Number of pores: number of objects • Pore characteristics of ellipses with the same second-moments as the pores - Orientation: angle between the major axis of the ellipse and the horizontal axis of the image ranging from −90 to 90° - Eccentricity: distance between the foci of the ellipse divided by the major axis length of the ellipse; ranges from 0 to 1, where eccentricity 0 means that the pore has spherical shape and eccentricity 1 that the pore is a line segment • Pore shape characteristics - Compactness factor: measurement for the complexity of a pore by comparing the contour of an object to its size [21] 4π  area cf ¼ 1  : perimeter2 Here, the pore area corresponds to the pore size and the perimeter to the contour, which refers to the number of pixels of an object having at least one background pixel as neighbor; the compactness factor for

elliptical objects is equal to 1  4πðπr 2Þ ¼ 0 and increases up to 1 for 2

ð2πrÞ

more complex shapes - Number of branching points per pore: number of skeleton pixels that lie on a line and have at least one orthogonal neighbor which is also a skeleton pixel whereas the other pixels are background - Number of endpoints per pore: number of skeleton points with only one other skeleton pixel in its neighborhood.

2.6. Statistical analysis Six pore characteristics were chosen for a statistical image analysis of whole film cross-sections from drum-to-air side. These pore characteristics were pore area fraction, pore size, number of pores and the three shape characteristics introduced in Section 2.5. The statistical analysis was conducted in the software R version 3.1.0. In order to have balanced data, film samples comprising 65 parts were shortened to 60 by removing five equidistant parts starting with the first image at the drum side. In a visual analysis, all calculated values for each pore characteristic at each position t were plotted with linear interpolation between the 12

points. Furthermore, the mean curves CðtÞ ¼ 1=12∑k¼1 CðkÞ ðtÞ over all 12 samples per film were compared, where C(k) is the pore characteristic C of the kth sample. The correlation structure between the two films was analyzed by Pearson's correlation coefficients ρ. For the correlation structure within a film, autocorrelation functions were studied. Here, we say that an autocorrelation of a random variable X is of order k if the value at position t depends on the k previous positions, in expression X t ¼ β0 þ β1 X t1 þ β2 X t2 þ … þ βk X tk þ εt ; where εt is independent and normally distributed noise with zero mean and constant variance. Here, the random variable Xt is the pore characteristic C(t) at position t and xt its observed value. The autocorrelation function of a stationary series {Xt}nt= 1 with translation-invariant mean μ and variance σ 2 at lag l∈f1; …; Lg; L b n is defined as ACF ðlÞ ¼

CovðX t ; X tþl Þ ¼ ρðX t ; X tþl Þ σ2

and can be estimated by d ðlÞ ¼ ACF

nl 1 ∑ ðxt  xn Þðxtþl  xn Þ; 2 ðn  lÞs t¼1

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

where xn denotes the sample mean and s2 the sample variance [22]. 95% pffiffiffi confidence bounds estimated by 2= n were also included in ACF plots. In order to statistically test whether there is a significant difference between the mean curves C of the SL and L film for each of the six pore characteristics, a three-way analysis of variance (ANOVA) was conducted. In the three-way ANOVA it was tested whether three factors A, B, and C have a significant influence on the observed values of these six pore characteristics. The first main factor A was chosen to be ‘type of film’ having two levels representing the SL film and the L film, respectively. The second factor B was ‘location of sample’ with two levels for sample strips taken from the side or from the middle of the respective piece of the film, see Fig. 1 again for the sample setup. This factor was included in the ANOVA to test whether the different samples can be seen as repeated observations from the same random variable. The last factor C was ‘position within sample’ with only three levels instead of 60. The average of parts 6–15 (drum side), 26–35 (middle), and 46– 55 (air side) in each sample were used as observations of the outcome variable. The choice of not including all parts in the ANOVA was made to reduce dependence between observations. Factor C was used to test whether the SL and L films have different layerings within their pore structure. In the ANOVA, the total variability in the used data ST was decomposed into the variability within different groups depending on the three chosen main factors, their interactions and the residual variability SR such that ST ¼ SA þ SB þ SC þ SðABÞ þ SðAC Þ þ SðBC Þ þ |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} main effects

2way interactions

SðABC Þ |fflfflfflfflffl{zfflfflfflfflffl}

þSR

3way interactions

assuming independent and normally distributed noise with mean zero and constant variance. It was tested whether the variability within the groups was greater than the residual variability. The test statistic is F(1,7)-distributed under the noise assumptions and the null hypothesis, which refers to the assumption that the respective factor has no effect on the respective pore characteristic [23]. We get 7 degrees of freedom for the residuals since we have 2 ⋅ 2 ⋅ 3 = 12 observations and 12- (2-1)(3 - 1) - (2 - 1) - 1 = 7. The multiplicity introduced by testing a family of six hypotheses on pore characteristics simultaneously was encountered using Bonferroni correction ([24], p.166) of the significance level α. Here, the ANOVA was conducted at a confidence level of α = 0.05, which was adjusted to αadj = α/6=8.33⋅ 10-3. 3. Results and discussion 3.1. Single image analysis In a preceding non-merged image analysis based on every other SEM image of each sample, all pore characteristics, but orientation and eccentricity were identified as relevant for further analysis on their impact on water permeability through the two studied EC/ HPC films. The pore characteristics based on ellipses with the same second-moment as the pores failed to capture any difference between the two films. For all samples from the SL and L films, the mean pore orientation was around 0 (± 4) and the mean eccentricity approximately 0.76 (± 0.04). The negligible values for the orientation and the relatively high value for the eccentricity indicate that the pores are elongated along the horizontal axis of the images. This characteristic of the pore structure can be explained by its formation during film spraying. 3.2. Merged image analysis In a merged image analysis along the cross-section, we found that all obtained curves C of each pore characteristic strikingly show a periodic change of pore structure within the cross-section even after averaging over the 12 respective samples per film (C). Fig. 4 gives an example of

155

such oscillating mean curves C for the mean pore size and number of branching points per pore for the SL and L films. Interestingly, the trajectories of a pair of mean curves ðCSL ; CL Þ occasionally have a peak at the same position t indicating some correlation between the two films. In fact, we found a strong correlation between the mean curves of the ^ ¼ 0:81Þ and a moderate correlation between the compactpore size ðρ ^ ¼ 0:23Þ . All other correlations were ness factor mean curves ðρ ^ b 0:1). Furthermore, we evaluated the correlation neglectable (0 b ρ within samples by studying autocorrelation functions as shown for the mean curves of the mean pore size and the number of branching points per pore in Fig. 4. Based on these plots, it was inferred that for all pore characteristics the autocorrelation is of order less than 10. Consequently, the spacing chosen for the third ANOVA factor ‘position within sample’ was sufficient to obtain at least linearly uncorrelated observations. For both films, the autocorrelation functions were in good agreement with the oscillating trajectories of the mean curves reflecting the periodic change in pore structure due to the film spraying process. As a result, it can be concluded from the between and within sample correlations that the pore size seems to be more affected by the spraying process than the pore shape. Furthermore, the L film seems to be more sensible to the nature of the spraying process than the SL film. Table 2 presents each test statistic and corrected p-values for the first factor ‘type of film’, in the respective ANOVA for the six tested pore characteristics. Table 2 additionally shows the overall mean values 60 1 60 ∑t¼1

CðtÞ and standard error for each pore characteristic and film. The obtained p-values show a high statistically significant difference between the SL and L films for pore area fraction and pore shape in terms of number of branching points per pore. Table 2 does not present the complete ANOVA results since the two other tested main factors, ‘location of sample’ and ‘position within sample’, do not seem to be significant and since no significant interactions between the main factors were found. This non-significant result for the second factor ‘location of sample’ strengthens the assumption that the films are homogeneous in each layer. The result that ‘position within sample’ does not appear to be a significant factor for a difference between the two films may speak for a stable film spraying process. In contrast to previous experimental findings on only a slightly higher film weight reduction after HPC leaching for the L film, the ANOVA showed a high statistically significant difference between the SL and L films for the pore area fraction (p =0.0005). A possible explanation can be that the L film has a larger number of pores with longer contours than the SL film, whose affect on pore area fraction is further enlarged by the applied image processing procedure dilating pores beyond their original size. However, it could also indicate that the L film is more robust during ultramicrotome cutting than the SL film. Last but not least, the volume fraction of the HPC-rich phases could differ for the two phase separating systems. Since the HPC-rich phase serves as a template for the formed pores visible in the film crosssections, a larger volume fraction of this phase in the L film would lead to greater pore area fraction. Leaching experiments cannot always expose such differences, since some HPC-rich domains may be trapped in the film by a surrounding EC-rich phase. The compactness factor (p = 0.04) showed a significant difference between the two films with a higher value for the L film indicating that the pores have more complex shapes than in the SL film. Since the two pore skeleton characteristics, number of branching points and ^ N 0:6 for both films) it was expectendpoints per pore, are correlated (ρ ed to get similar results for them. Nonetheless, the p-value for the number of branching points per pore (p = 0.0006) is smaller than for the number of endpoints per pore (p = 0.0095). Based on the larger values for the L film, we concluded that the pores are not only more branched but also more extended in the L film than in the SL film. The higher permeability measurements for the L film suggest that the more extended and complex shapes result in a better connected pore system in the L film than in the SL film. Hence, pore connectivity

156

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

Fig. 4. Left: mean curves from drum-to-air side through SL (dashed) and L (solid) films for mean pore size in μm2 (top) and number of branching points per pore (bottom). A clear difference between the SL and L curves can be detected as well as a stronger amplitude of the oscillating curves for the L film. Right: autocorrelation functions (ACF) for the mean curves of the SL film (upper) and L film (lower) for mean pore size (top) and number of branching points per pore (bottom) including 95% confidence bounds (dashed).

seems to be the most important difference in pore structure between the two films. In Fig. 5, it is schematically shown how a more branched and complex pore shape in cross-sections could originate from a better connectivity between pore channels (Example B). The two examples shown would induce about the same amount of leached HPC under the prerequisite that the HPC-rich domains forming the channels would be connected to a water-exposed surface. However, the dead end resulting from a lower degree of connectivity and, hence, a simpler

Table 2 Sample mean (standard error) of six pore characteristics for the SL and L films, as well as ANOVA results of factor ‘type of film’, presenting the test statistic (F(1,7)-value) and corrected p-values (b0.001 highly significant, b0.05 significant). Highly significant results are highlighted in bold. Pore characteristic

SL

L

F(1,7)

p-Value

Pore area fraction Pore size in μm2 Pore number Compactness factor Branching points Endpoints

0.253 (0.012) 0.315 (0.017) 56.37 (2.041) 0.264 (0.009) 0.229 (0.035) 2.217 (0.030)

0.299 (0.006) 0.338 (0.022) 63.50 (3.840) 0.288 (0.010) 0.408 (0.036) 2.351 (0.062)

65.506 3.959 26.471 14.477 60.356 24.839

0.0005 0.5214 0.0080 0.0400 0.0006 0.0095

cross-sectional pore shape in Example A would lead to a lower permeability since molecules entering the dead end have less possibilities to leave that pore. Relating pore connectivity further to percolation theory, the SL film might be closer to the percolation threshold than the L film as the threshold value may change with HPC viscosity grade. In a threedimensional validation study, it will be possible to draw conclusions on the number of dead ends in the pore structure and its relationship to the pore skeleton characteristics from two-dimensional image analysis. It should be noted that we analyzed one sample of an EC/HPC film available with even lower HPC viscosity grade and measured permeability than for the SL film. The SEM cross-section images of this film showed a well separated spherical pore structure. We found a coherent difference between all three films for pore area fraction and the introduced shape characteristics. Since there was only one such sample available, the results are only indicative and not presented above. Furthermore, it should be mentioned that pronounced differences in pore size have been observed by SEM at the outermost layers of EC/HPC-films, with the air side generally having smaller pores than the internal structure [10,25]. Such pore heterogeneities were also observed in the SEM images for the films used in this study, but the manual image cropping of the outermost ends of the film excluded

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

157

Fig. 5. Schematic illustration of pore channels inside a film with cross-section images indicated by transparent rectangles. Two different connectivity examples, A and B, are presented. Compared to a simpler, elliptical cross-sectional pore shape shown in Example A of a system with two unconnected channels, a more branched and complex pore shape originated from a better connected system of pore channels depicted in Example B.

these layers from the analysis. This selection was done since more advanced and meticulous image analysis on the outmost layers is needed. 4. Conclusions We characterized cross-sections of porous EC/HPC films used for controlled drug release formulations. For such characterizations of porous materials, we developed a suitable image processing procedure based on standard tools from image analysis. Here, image merging was important for obtaining non-overlapping and tight subsections along cross-sections. In this way, a complete characterization of the pore structure with regard to film depth was achieved, which may be applicable for other porous materials as well. The pore characterization of two films with different HPC molecular weight was used to study how HPC viscosity grades may affect the pore structure and, hence, mass transport through EC/HPC films. The film with higher molecular weight (L) was about three times more permeable than the film with lower molecular weight (SL), even though these films seemed similar and showed comparable weight reduction after HPC leaching. Applying our image processing procedure and conducting statistical analysis, we found a high statistically significant difference in pore shape in the SL and L films. Concerning the pore shape, we discovered that the skeleton of a pore describes its complexity well. In particular, the number of branching points per pore indicated that the complex pores in the L film are more extended than in the SL film. This main conclusion is in agreement with a common hypothesis on permeability in porous materials stating that larger and more complex pores are favorable for high pore connectivity and, hence, mass transport. That is why it can be inferred that HPC molecular weight may be useful for predicting and controlling mass transport through EC/HPC films. However, further studies should be conducted including the outmost film layers, replicates of each type of film and a larger variety of considered HPC molecular weights. We want to link our findings on pore shape and autocorrelation structure through cross-sections to three-dimensional data obtained not only from free films, but also from coated pellets. A study on pellet coatings with different molecular weight of EC [26] suggests that results from free films may be transferable to drug release mechanisms of coated dosage forms. It will be interesting to confirm this hypothesis in future studies. For this purpose, we first aim to reconstruct the pore structures observed in free films. Having constructed a model for the pore structure, we can conduct a simulation study on mass transport through simulated pore structures. Such three-dimensional analyses could bring forth a greater understanding of pore connectivity and its relationship to mass transport through porous controlled drug

release films. Furthermore, we hope to show that already from a twodimensional analysis as presented in this paper, conclusions about the three-dimensional pore structure can be drawn — a technique that could also be useful for other porous systems within a large variety of applications. Acknowledgment This work is part of the VINN Excellence Centre SuMo BIOMATERIALS and has mainly been financed by the Swedish Governmental Agency for Innovations Systems, VINNOVA. In addition, the financial support from the Knut and Alice Wallenberg Foundation, KAW (KAW 2012.0067) , and the Swedish Foundation for Strategic Research, SSF (SSF AM13-0066), is highly appreciated. The authors would also like to thank Mats Rudemo, Anne-Marie Hermansson and Stefan Gustavsson from Chalmers University of Technology, Mats Stading from SP Food and Bioscience as well as Mariagrazia Marucci and Christian von Corswant from AstraZeneca R&D Mölndal for inspiring discussions and helpful advice. The authors are grateful to Sofie Sandhagen (former Olsson) for experimental method development and film preparation during her Master project at the Department of Chemical and Biological Engineering, Chalmers University of Technology and SIK — The Swedish Institute for Food and Biotechnology (currently the Department of Chemistry and Chemical Engineering and SP Food and Bioscience). The authors also thank the reviewers for their valuable comments. References [1] F. Siepmann, J. Siepmann, M. Walther, R.J. MacRae, R. Bodmeier, Polymer blends for controlled release coatings, J. Control. Release 125 (1) (2008) 1–15, http://dx.doi. org/10.1016/j.jconrel.2007.09.012. [2] P.J. Dees, J. Polderman, Mercury porosimetry in pharmaceutical technology, Powder Technol. 29 (1) (1981) 187–197, http://dx.doi.org/10.1023/A:1009630015598. [3] C. Boissier, F. Feidt, L. Nordstierna, Study of pharmaceutical coatings by means of NMR cryoporometry and SEM image analysis, J. Pharm. Sci. 101 (7) (2012) 2512–2522, http://dx.doi.org/10.1002/jps.23160. [4] D. Benjamini, J.J. Elsner, M. Zilberman, U. Nevo, Pore size distribution of bioresorbable films using a 3-D diffusion NMR method, Acta Biomater. 10 (6) (2014) 2762–2768, http://dx.doi.org/10.1016/j.actbio.2014.02.014. [5] R.A. Siegel, Porous Systems, in: Fundamentals and Applications of Controlled Release Drug Delivery, Springer, New York, 2012 29–251. [6] M. Marucci, J. Hjärtstram, G. Ragnarsson, F. Iselau, A. Axelsson, Coated formulations: new insights into the release mechanism and changes in the film properties with a novel release cell, J. Control. Release 136 (2009) 206–212, http://dx.doi.org/10. 1016/j.jconrel.2009.02.017. [7] R. Bergman, L.O. Sundelöf, Diffusion transport and thermodynamic properties in concentrated water solutions of hydroxypropyl cellulose at temperatures up to phase separation, Eur. Polym. J. 13 (11) (1977) 881–889, http://dx.doi.org/10. 1016/0014-3057(77)90060-X.

158

H. Häbel et al. / Journal of Controlled Release 222 (2016) 151–158

[8] M. Marucci, J. Arnehed, A. Jarke, H. Matic, M. Nicholas, C. Boissier, C. von Corswant, Effect of the manufacturing conditions on the structure and permeability of polymer films intended for coating undergoing phase separation, Eur. J. Pharm. Biopharm. 83 (2) (2013) 301–306, http://dx.doi.org/10.1016/j.ejpb.2012.09.020. [9] P. Sakellariou, R.C. Rowe, Interactions in cellulose derivative oral drug delivery, Prog. Polym. Sci. 20 (1995) 889–942, http://dx.doi.org/10.1016/0079-6700(95)00008-4. [10] H. Andersson, J. Hjärtstam, M. Stading, C. von Corswant, A. Larsson, Effects of molecular weight on permeability and microstructure of mixed ethyl-hydroxypropylcellulose films, Eur. J. Pharm. Sci. 48 (2013) 240–248, http://dx.doi.org/10.1016/j. ejps.2012.11.003. [11] M. Larsson, J. Hjärtstam, J. Berndtsson, M. Stading, A. Larsson, Effect of ethanol on the water permeability of controlled release films composed of ethyl cellulose and hydroxypropyl cellulose, Eur. J. Pharm. Biopharm. 76 (3) (2010) 428–432, http://dx. doi.org/10.1016/j.ejpb.2010.09.007. [12] E.L. Cussler, Diffusion: Mass Transfer in Fluid Systems, second ed. Cambridge University Press, Cambridge, 1997. [13] MATLAB, Version 8.2.701 (R2013b), The MathWorks, Inc., Natick, USA, 2013 (URL http://www.mathworks.com/). [14] M.A. Fischler, R.C. Bolles, Random sample consensus: a paradigm for model fitting with apphcatlons to image analysis and automated cartography, Commun. ACM 24 (6) (1981) 381–395, http://dx.doi.org/10.1145/358669.358692. [15] C. Harris, M. Stephens, A combined corner and edge detector, Procedings of the Alvey Vision Conference 1988 1988, pp. 231–236, http://dx.doi.org/10.5244/C.2.23. [16] D.G. Lowe, Distinctive image features from scale-invariant keypoints, Int. J. Comput. Vis. 60 (2) (2004) 91–110, http://dx.doi.org/10.1023/B:VISI.0000029664.99615.94. [17] M. Brown, D.G. Lowe, Automatic panoramic image stitching using invariant features, Int. J. Comput. Vis. 74 (1) (2006) 59–73, http://dx.doi.org/10.1007/s11263-0060002-3.

[18] E. Schuster, J. Eckardt, A.M. Hermansson, A. Larsson, N. Lorén, A. Altskär, A. Ström, Microstructural, mechanical and mass transport properties of isotropic and capillary alginate gels, Soft Matter 10 (2) (2014) 357–366, http://dx.doi.org/10.1039/ c3sm52285g. [19] N. Otsu, A threshold selection method from gray-level histograms, IEEE Trans. Syst. Man Cybern. Syst. 9 (1) (1979) 62–66, http://dx.doi.org/10.1109/TSMC.1979. 4310076. [20] L. Lam, S.W. Lee, Thinning methodologies — a comprehensive survey, IEEE Trans. Pattern Anal. Mach. Intell. 14 (9) (1992) 869–885, http://dx.doi.org/10.1109/34. 161346. [21] R.M. Rangayyan, N.M. El-Faramawy, Measures of acutance and shape for classification of breast tumors, IEEE Trans. Med. Imaging 16 (6) (1997) 799–810, http://dx. doi.org/10.1109/42.650876. [22] P.M.T. Broersen, Periodogram and lagged product autocorrelation, Automatic Autocorrelation and Spectral Analysis, Springer, London 2006, pp. 29–57. [23] B.H. Cohan, Two-way ANOVA, Explaining Psychological Statistics, John Wiley & Sons, Hoboken 2008, pp. 425–486. [24] L. Wasserman, All of Statistics: A Concise Course in Statistical Inference, Springer, New York, 2004. [25] T. Gebäck, M. Marucci, C. Boissier, J. Arnehed, A. Heintz, Investigation of the effect of the tortuous pore structure on water diffusion through a polymer film using lattice boltzmann simulations, J. Phys. Chem. B 119 (2015) 5220–5227, http://dx.doi.org/ 10.1021/acs.jpcb.5b01953. [26] M. Marucci, H. Andersson, J. Hjärtstam, G. Stevenson, J. Baderstedt, M. Stading, A. Larsson, C. von Corswant, New insights on how to adjust the release profile from coated pellets by varying the molecular weight of ethyl cellulose in the coating film, Int. J. Pharm. 458 (2013) 218–223, http://dx.doi.org/10.1016/j.ijpharm.2013. 09.016.