In-Process Control Assay of Pharmaceutical Microtablets Using

Subscriber access provided by UniSA Library

Article

In-process control assay of pharmaceutical microtablets using hyperspectral imaging coupled with multivariate analysis Lalit Mohan Kandpal, Byoung-Kwan Cho, Nishanth Gopinathanc, and Jagdish Tewari Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.6b02969 • Publication Date (Web): 12 Oct 2016 Downloaded from http://pubs.acs.org on October 17, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

In-process control assay of pharmaceutical microtablets using hyperspectral imaging coupled with multivariate analysis Lalit Mohan Kandpal, † Byoung-Kwan Cho,*† Nishanth Gopinathanc, § and Jagdish Tewari*¶ †,*†

Department of Biosystems Machinery Engineering, College of Agricultural and Life Science, Chungnam National University, 99 Daehak-ro, Yuseong-gu, Daejeon 305-764, South Korea § Formulation Development, Biogen, Cambridge, MA, USA. *¶ Process Analytical Technology, Analytical Development, Biogen, Cambridge, MA, USA. ABSTRACT: Monitoring the amount of active pharmaceutical ingredient (API) in finished dosage form is important to ensure the content uniformity of the product. In this report, we summarize the development and validation of a hyperspectral imaging (HSI) technique for rapid in-line prediction of the active pharmaceutical ingredient (API) in microtablets with concentrations varying from 60% to 130% API (w/w). The tablet spectra of different API concentrations were collected in-line using an HSI system within the visible/near-infrared (Vis/NIR; 400–1000 nm) and short-wave infrared (SWIR; 1100–2500 nm) regions. The ability of the HSI technique to predict the API concentration in the tablet samples was validated against a reference high-performance liquid chromatography (HPLC) method. The prediction efficiency of two different types of multivariate data modeling methods, i.e., partial leastsquares regression (PLSR) and principle component regression (PCR), were compared. The prediction ability of the regression models was cross-validated against results generated with the reference HPLC method. The results obtained from the PLSR models showed reliable performance for predicting the API concentration in SWIR region. The highest coefficient of determination ( p) was 0.96, associated with the lowest predicted error and bias of 4.45% and -0.35%, respectively. Furthermore, the concentrationmapped images of PLSR and PCR models were used to visually characterize the API concentration present on the tablet surface. Based on these results, we suggest that HSI measurement combined with multivariate data analysis and chemical imaging could be implemented in the production environment for rapid in-line determination of pharmaceutical product quality.

The major profit source of the growing pharmaceutical industry is development of innovative new drug products. However, finalizing any innovative drug product is a complex process that requires an average of 12 years, during which time research expenditures accumulate and clinical trials and quality assurance tests are carried out.1 Once the drug is finalized by quality testing, it can be launched in the marketplace and become available for patient use. However, it is necessary to ensure that the intermediate or finished product conforms to the necessary standards and specifications. Thus, the pharmaceutical industry requires the use of advanced analytical techniques during the manufacturing process to evaluate product quality. Identification of active pharmaceutical ingredients (APIs) during different stages of the production process is an essential quality attribute that can be used to confirm the content uniformity of solid dosage forms.2 Because the final product must have a specific API concentration to achieve the desired performance and therapeutic effects,3,4 low API concentrations may lead to large variability in dosage forms, affecting the dissolution properties of the drug product. In the pharmaceutical industry, in-process control (IPC) strategies are widely performed at regular intervals to monitor the production of intermediate or API components. IPC refers to the qualitative and quantitative analysis of raw material before the manufacturing process completed. After IPC checks, if necessary, the defective batch can be improved by adjusting the API concentration or formulation. An important criterion of IPC testing is

speed; the production chemist requires a rapid data turnaround during and after manufacturing.5 Therefore, it is essential to apply rapid analytical methods to scale-up the IPC process. Currently, many offline technologies and strategies (e.g., high-performance liquid chromatography [HPLC], mass spectrometry [MS], and ultraviolet [UV] spectrophotometry) are used in IPC for quality analysis of pharmaceutical medications.5-7 However, such technologies are generally used for API monitoring and are time consuming, tedious, and often destructive in nature.8,9 Hence, more rapid and accurate techniques are needed for real-time testing and validation of the final product. Moreover, rapid control of the manufacturing process using process analytical technology (PAT), promoted by the Food and Drug Administration (FDA), provides important opportunities for improving our understanding and control of the process, while decreasing safety risks associated with sampling and testing.10,11 An alternative to traditional methods is hyperspectral imaging (HSI), a spectroscopy technique that offers a rapid, accurate, and nondestructive quality control method for drug products. HSI is a well-known PAT tool that can be used for in-process quality monitoring of biological and biomedical materials in the pharmaceutical sector. Furthermore, HSI chemical imaging (HSI-CI) is a powerful technique that has several advantages over traditional spectroscopy techniques. For example, HSI-CI provides sufficient information regarding the chemical concentration present on the sample surface, based on spatial and spectral features.12,13 Previous approaches

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

have already proven the potential of the HSI-CI method for quality monitoring of pharmaceutical materials. These approaches have used the HSI-CI technique for monitoring of API contents in tablet and powder samples.14-17 However, most of these approaches have utilized HSI or near-infrared (NIR) spectroscopy for API determination in single and large tablet form, and no studies have reported the use of the HSI technique for IPC assay determination of API concentrations in bulk microtablets. Microtablet is the tablet with a diameter equal to or less than 3 mm. They have several advantages over larger tablets, such as relatively easy to be manufactured, convenient to use, immediate release, less risk of dose dumping, better size uniformity, and high mechanical strength.18 Pharmaceutical compounds typically have unique spectral signatures of the drug API compositions, particularly in the shortwave-infrared (SWIR) region from 1000 to 2500 nm.4 Therefore, HSI in the SWIR region allows access to information regarding organic compounds that have overtone and vibrations. Extracting the relevant information from the HSI data to build a calibration model is a challenging task that must be overcome. The major problem observed with HSI is interpretation and evaluation of large amounts of data consisting of several variables and complex spectra. The complexity of the spectra makes correlation with quantitative properties difficult without preprocessing and subsequent chemometric analysis. Hence, the use of multivariate analysis methods is crucial for identification of the relationship between the spectral characteristics and the material properties of interest. For this purpose, several chemometric and preprocessing methods have been tested in order to obtain quantitative and qualitative information of pharmaceutical materials.12, 17,19-22 Chemometric analysis yields predictive models that enable the at-line or online rapid analysis of materials that were traditionally analyzed using time-consuming analytical techniques, such as HPLC. The objective of this study was to investigate the potential applicability of HSI IPC assays employed with high drug load for small and thin microtablet manufacturing, which accounts for many bulk tablet samples. The main goal was to develop an IPC assay that could be performed at-line on the manufacturing floor without any sample preparation. We also aimed to develop multivariate analysis and image processing models for API prediction and visualization of content distribution in tablet samples.

EXPERIMENTAL SECTION Microtablets. The microtablets used during this study were kindly supplied by Biogen Idec, Cambridge, United States (Figure1). The tablet samples were produced by compressing the API and excipients powders. The API is assigned as API instead of its original chemical name to protect the company law. Tablets having diameter of 2.18 mm (single tablet) were used during the experiment. The common excipient/placebo ingredients were microcrystalline cellulose (MCC), silicon dioxide, and mannitol. In addition, magnesium stearate (MgSt) is used as a coating material. The tablets of eight batches were produced by targeting the API concentration range from 60% to 130% (w/w), by weighting various amounts of API and excipients compounds. The tablets were weighted after preparation, and the results indicated that the amount of material that adhered to the punches was insignificant. The detailed information about tablet preparation is summarized in Table 1.

Page 2 of 9

Figure 1. Arrangement of tablet samples for HSI spectra collection. Table.1. Summary of the approximate concentration of API and other compounds for tablet preparation. %(w/w)

API

Placebo (g)

MgSt (g)

Target (%w/w) API

Placebo

60

117

184.5

1.5

39

61

70

136.5

165

1.5

45

55

80

156

145.5

1.5

51

49

90

175.5

126

1.5

58

42

100

195

103. 5

1.5

65

35

110

2.14.5

87

1.5

71

29

120

234

87

1.5

73

27

130

253.5

48

1.5

84

16

HPLC. HPLC analysis of microtablets was carried out on a Waters Alliance 2690 instrument with a photodiode array detector (Milford, MA, USA). The procedure was used as a reference method to determine the API content of tablet samples. Weighed portions of microtablets were transferred into 500-mL volumetric flasks, and 300 mL MeOH was added to each volumetric flask. Samples were prepared in 50/50 (v/v) phosphate buffer/MeOH and filtered with 0.7-µm nylon syringe filters. Chromatographic separation was performed at a column temperature of 30°C, autosampler temperature of 5°C, and flow rate of 1.2 mL/min on a 2.6-µm Kinetex C18 column (4.6 × 150 mm; Phenomenex, Torrance, CA, USA). A 5-µL injection volume was loaded onto the column and analyzed over 8 min. UV absorption was measured at 223 nm, and Empower Pro was used to analyze the chromatograms. HSI instrumentation and image acquisition. Two types of laboratory based line-scan hyperspectral systems, i.e., visible/NIR (Vis/NIR) and SWIR, were used for microtablet measurement. The detailed description about both system are as follows. The Vis/NIR HSI system was composed of EMCCD detector (MegaLuca R; ANDOR Technology, USA), imaging spectrograph (Headwall Photonics, Fitchburg, MA, USA), objective lens (focal length of 28 mm f/1.4).The Vis/NIR camera captured the image of 128 spectral bands from 4001000 nm wavelength range with 1004 × 1002 total pixels (Detector size) and frame rate of 12.4 Hz. Whereas, the SWIR system was comprised of a mercury cadmium telluride (MCT) detector (Model: Xeva-2.5-320; Xenics, Belgium) and objective lens (focal length of 25 mm f/1.4). The SWIR camera captured the image of 208 bands from 1000 to 2500 nm with 320 × 256 total pixels and frame rate of 100Hz. The SWIR


Page 3 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


system operated at temperature between 0 to 50ºC and cooling down time of camera was < 300 sec. Both system also composed of light source (halogentungsten lamps) to illuminate the sample, translation stage for moving the sample, and DC motor to control the speed of translation stage. The control program for HSI systems were developed using Microsoft (MS) Visual Basic (Version 6.0) software operated in MS Windows. The basic components of HSI systems are shown in Figure 2.

Figure 2. Schematic of hyperspectral imaging system of tablet samples. The tablet samples of different API concentrations (shown in Figure 1) were placed into a sample holder plate (800 mm length). A single holder consisted of approximately 15 tablet samples of particular concentration. Before HSI image acquisition of the tablet samples, the system parameters were adjusted: the sample increment was adjusted as 0.2 mm/scan, the exposure time was adjusted as 7500 micros and the CCD cooling temperature of 200 K. Further, the sample plate was transferred to the translation stage and scanned line by line using both HSI systems. The DC motor moved the translation stage towards the HSI camera. The HSI camera mounted over the stage began to acquire the image as the tablet samples entered in the camera field of view (FOV). The samples were scanned line-by-line under both HSI systems in the wavelength range from 400 to 2500 nm (VIS/NIR and SWIR) with a spatial resolution of 12 microns/pixel. Finally the acquired hyperspectral images were saved in a three-dimensional (3D) format containing two spatial dimensions (x and y) and a spectral dimension (λ). Image correction and data preprocessing. The white and dark reference images were acquired to correct original sample HSI images from the dark current of the camera. To calibrate the images, a white image (100% reflectance) was obtained with a white ceramic Teflon sheet, and a dark image (0% reflectance) was obtained by turning off the light source and covering the lens. Finally, the relative reflectance was calculated by applying the following equation.

=

(1)

where is the original image, is the dark image, is the white image, and is the corrected image. Furthermore, a region of interest (ROI) step was performed on the HSI image to extract the spectral information for further analysis. The ROI pixel selection was computed manually, and the extracted pixel spectra from tablets at each API concentration were averaged before further analysis. Therefore, a total of 400 spectra from eight API concentrations (50 spectra × 8) were extracted and used for analysis. The hyperspectral data were influenced by undesirable noise affecting the spectral features and prediction efficiency of the model. To avoid the influence of the noise from the

spectra, spectral pretreatment is crucial. In this study, the ROI selected data were pretreated using several preprocessing methods, including multiplicative scatter correction (MSC), standard normal variate (SNV), and Salvitzky-Goaly (SG) filtering. MSC and SNV are the most common preprocessing methods used to correct the background offset, slope, and scattering effect from the data.22 However, the Salvitzky-Golay filters (SG-first and second derivatives) were used to reduce the additive effects and resolve the overlapping peaks of the spectra.23,24 After the preprocessing steps were completed, the spectral data for the tablet samples were analyzed using multivariate analysis methods, including principle component analysis (PCA), principle component regression (PCR), and partial least square regression (PLSR) with MATLAB software. PCA model. PCA is a well-known technique most frequently applied to spectroscopic data, for data dimension reduction, for identifying data patterns, and for computing latent variables (LVs) for regression models.25,26 The score of the PCA is the projection of the observations/samples in the principle component (PCs). The PCA score can be used to visualize multivariate data with scatter plots in order to interpret the relationships among the observations/samples. The loading plot is used for interpreting relationships among variables. The general PCA model is expressed as.

X= TP + E (2) where X is the original matrix; T and P are the score and loading matrix, respectively; and E is the residual matrix. In this study, the X matrix was composed of the original spectral data for the tablet samples with different API concentrations. PCR and PLSR model. The PCR and PLSR models were applied to build prediction models for tablet APIs. PCR and PLSR are generally used when there are a large number of predictive variables and when those predictors are highly correlated or collinear. PCR is a combination of PCA and multivariate linear regression (MLR). The first step of PCR is to perform PCA and decompose the spectral data by applying equation 2. Next, the optimal number of principle components/latent variables (LVs) obtained from PCA can be used in an MLR model to perform PCR.27 The PLSR model uses the X and Y matrices of dependent and independent variables, respectively. The analysis determines the linear relationship between the X and Y variables and then predicts the properties of the Y variable.28 The model is defined as. X= TP + E

(3)

Y= UQ + E (4) where X and Y are the independent and dependent variables, respectively; T and U are the score matrices; P and Q are the loading matrices of X and Y, respectively; and E is the error matrix. In this study, the X matrix consisted of the spectral data for the tablet samples, while the Y matrix consisted of eight API concentration values ranging from 60% to 130%. The entire set of data (X and Y matrices) was split into calibration and validation groups consisting of 280 samples (35 samples for each concentration) and 120 samples (15 samples for each concentration), respectively. Furthermore, regression models (PLSR and PCR) were developed using the calibration set and tested using the validation set. One important step during the design of the calibration model is determination of the number of LVs or factors that need to be considered. The number of



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

LVs can have a significant impact on the model. Improper selection of LVs can cause “overfitting” or “underfitting”, leading to spectral noise in the regression model and suppression of spectral information and model interpretation. Therefore, to prevent the model from having these faults, it is critical to select an optimal number of LVs. In this study, the optimum number of LVs was selected by using the lowest value of the root mean square (RMS) during the cross-validation (CV) process by applying equation 5. =

1

∑1

2

(5)

are the actual and predicted API concenwhere, " and tration values, respectively, and n is the number of predictions. Furthermore, the prediction efficiency of the PLSR model was assessed by calculating the following parameters; the coefficient of determination ( c, cv and p, respectively), and standard error of calibration, CV and prediction (RMSEC, RMSE*+ and RMSEP , and bias values. However, a good model provides high R values with low error and bias values.

∑. - 1 -/0,- ,

(6)

∑. 2 1 -/0,- ,

3

4

54

789

4

5

∑5"64" "

∑5"64" "

(7)

Page 4 of 9

Figure 3. SWIR spectra. (A) Original spectra. (B) SNV preprocessed spectra. The raw spectra for all tablets showed similar patterns for different API concentration, and clear spectral differences were not observed. However, the average spectra clearly demonstrated variations in intensity between each API concentration in tablets samples (Figure 4). The average spectra of samples in the SWIR region between 1000 and 2100 nm showed major differences in reflectance intensities. These intensity differences may have occurred due to the increased API concentration in the samples. The differences were particularly apparent at around 1360 and 1692 nm, where the signals could be attributed to the C--H combination and C--H stretch first overtone of the methyl (CH; ) group. In contrast, the regions at 1160, 1170 and 1950 nm belonged to the C==O stretch fourth overtone, C-H second overtone and C==O second overtones.28 These groups were attributed to the overall overtones of the API structure in comparison with pure API spectra, as shown in Figure 4A. In addition, significant changes in absorbance at various regions were observed, being to the varying API concentration corresponding first derivate spectra (Figure 4). In the VIS/NIR region (400–1000 nm) of HSI there was no significant spectral characteristics relating to the API content; therefore, we represented only SWIR spectra that enable chemical features of the tablet samples.

(8)

RESULTS AND DISCUSSION Spectral interpretation. Figure 3A,B shows the original spectral profiles and SNV preprocessed spectral profiles of tablet data with different API concentration. The original spectra of tablet samples contained various overlapping peaks and noise generated from the camera. In contrast, after applying SNV preprocessing, these effects were largely removed, and peaks were highly enhanced.

Figure 4. SWIR spectra of microtablet samples. (A) Average spectra (B) SG-1st derivate pretreated spectra applied with 3rd order polynomial for 5-point moving window size. PCA model for data visualization. First, PCA was carried out using the preprocessed data to confirm the ability of the model to discriminate between tablets of different API concentrations. The scatter plots of tablet samples obtained from PCA showed clear clustering of the tablet data according to the API concentration (Figure 5). The first three PCs captured 94% of total X-variance in the tablet samples. The model was capable of isolating tablets with higher API concentrations (90–130% [w/w] API tablets) clearly, whereas samples


Page 5 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


with lower API concentrations (60–80% [w/w] API tablets) were overlapped. Although the smaller API quantity was difficult to separate by PCA, it provided real-time information regarding the appropriate API concentration and was added to the tablet samples for visualization of the sample pattern.

Figure 5. Principle component analysis of tablet samples. PCR and PLSR models for API prediction. The PCR and PLSR models were used to predict the API concentrations of tablet samples in both HSI regions. Table 2 summarizes the overall chemometric prediction results of both HSI data using various pre-processing methods. Based on the results, the SWIR performed superior than Vis/NIR HSI for API prediction. This perhaps due to the SWIR region comprised of the broad bands and fundamental vibrations information, associated with API structure of the tablet samples. The prediction

ability of PLSR and PCR models in SWIR region is presented in Figure 6A and 6B. The prediction plots showed a linear relationship between the actual and predicted API concentrations. In SWIR HSI region the PLSR predicted highest coefficient of determination (R p) of 0.96, with minimum RMSEP and bias values of 4.45% and -0.35%, respectively, using nonprocessed spectral data. Similarly, the PCR model also predicted higher < values of 0.95 with minimum RMSEP and bias values of 5.45% and -0.35%, respectively, which was also obtained from non-processed spectral data. In addition, a slight decrease in predictive performance was observed in Vis/NIR HSI region. In this region, the PLSR predicted highest p and RMSEP values were 0.93 and 5.54% respectively, using SG-1st derivate pretreated spectral data. The optimum number of PLSR factors were determined by selecting the lowest RMSE*+ value. The optimum six factors shown in Figure 7A provided the lowest RMSC errors and high prediction accuracy in the cross-validation (CV) method for both models (PLSR and PCR) in SWIR region. The beta coefficient plot obtained from the PLSR model also showed important bands mostly related to the methyl (CH; ) and carbonyl (C=O) groups of the API structure of tablet samples (Figure 7B). In Vis/NIR region the minimum 7 PLS factors provided the lowest RMSC values. Furthermore, the 15 samples from different lots were collected to predict IPC assay using HPLC and HSI data to validate the HSI method. The validation results from HPLC showed < of 0.99 and %RSD of 2.1 respectively, for API prediction.

Figure 6. Predicted API concentration of tablet samples in SWIR region. (A) PLSR model. (B) PCR model.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 9

. Figure 7. (A) Optimum number of latent variables selected by cross-validation method. (B) Beta coefficient plot of second derivate preprocessing.

Table.2. Predicted API concentration of microtablet samples form PLSR and PCR model. Region

Model/Processing

LVsa

R calb

R vald

RMSE=> (%)e

R p f

PLSR/None

6

0.97

h

3.53

0.96

3.92

4

PLSR/MSC

i

0.95

4.83

0.94

PLSR/SG-1

j

4

0.95

4.60

PLSR/SG-2

k

4

0.97

5

PCR/None PLSR/None

PLSR/SNV SWIR

Vis/NIR

RMSEC (%)c

RMSEP (%)g

Bias (%)

0.96

4.45

-0.35

5.32

0.94

5.34

-0.69

0.95

5.09

0.93

5.85

-1.43

3.79

0.96

4.18

0.96

4.49

-0.91

0.95

4.95

0.94

5.48

0.92

6.29

1.79

6

0.96

4.10

0.96

4.66

0.95

5.45

-0.35

7

0.94

4.68

0.94

5.57

0.93

5.84

-3.04

PLSR/MSC

8

0.93

5.64

0.91

6.85

0.91

6.71

-3.23

PLSR/SG-1

5

0.95

5.06

0.91

6.78

0.90

7.25

-2.94

PCR/SG-1 7 0.95 5.01 0.94 5.58 0.93 5.54 -3.32 b,d,f c,e,g Number of latent variables. Coefficient of determination for calibration, cross-validation and prediction. Root mean square error of calibration, cross-validation and predication. fStandard normal variate. gMultiplicative scatter correction. h,iSalvitzky-Golay first and second derivative. a


Page 7 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 8. Concentration map of tablet API. (A) RGB image. (B) HSI image. (C) Masking image. (D) PLSR image (Vis/NIR: 4001000 nm). (E) PCR image (SWIR: 1100-2500 nm). (F) PLSR image (SWIR: 1100-2500 nm). API distribution-mapped images. The concentrationmapped images of tablets with different API concentrations are depicted in Figure 8. Before the development of concentration-mapped images, the background was eliminated from the samples using a simple threshold method (Figure 8C). Further concentration images were constructed using the beta coefficient values extracted from the PCR and PLSR models. These images not only provided the spatial distribution of the predicted API concentrations into the tablet surface but were also useful to ensure that the specific API concentration was added in the final dosage form. The color scale from blue to red shows the API concentration distribution ranging from 0% to 130%. Pure API was used as a reference compound to ensure that the pixels corresponded to the tablet API, and the remaining pixels corresponded to the excipient compounds. The color bar represents the intensity of the API concentration. From the resulting images, we noticed that the API pixels had a more intense red color than those of the excipients. Based on this observation, we concluded that the red pixels located on the tablet surface were related to the API compound. The concentration-mapped images of the PLSR models from both HSI instruments (VIS/NIR and SWIR) could be used to easily visualize the API distribution on the overall tablet surface (Figure 8D and E). However, the visualization and characterization were difficult for tables with lower API concentrations in the PCR model. The PCR model was unable to detect any API pixel in tablets with lower concentrations (60–80% [w/w] API tablets), whereas tablets with higher concentrations (90–130% [w/w] API tablets) were easily characterized (Figure 8E) by

the model. In addition, some abnormalities, including dead pixels and blank pixels in the SWIR region, could also have influenced the image performance. Although it is difficult to visualize the API concentration using via traditional methods, the concentration-mapped images generated from HSI and image processing techniques provided an effective method for visual observation of the API concentration in the final dosage form.

CONCLUSIONS Quality control of pharmaceutical APIs is linked to content uniformity and the dissolution properties of the solid dosage forms and therefore plays an important role in the drug development process. The pharmaceutical API should be homogeneous with specified limits in solid dosage forms. In this current approach, the HSI technique was utilized for in-process testing of pharmaceutical API in microtablets. Tablet samples containing different API concentrations were investigated through HSI instruments from the VIS to the SWIR region (400–2500 nm). Three different chemometric models, i.e., PCA, PCR, and PLSR, were evaluated for analysis of API contents of tablet samples. A comparative study showed that both PCR and PLSR models had the best prediction ability for tablet API contents with strong value of greater than 0.90 and the lowest SEP of 4.45%. In addition, the concentration-mapped images of tablet APIs provided a visual characterization of tablets with different API concentrations and the distribution pattern of API component on the tablet surface. The distribution pattern of API was clearly visualized on the entire tablet surface using the PLSR model; however, the image generated from the PCR model was unable to clearly visualize the lower API concentration range from 60% to 80% (w/w) API. The overall results demonstrated that the HSI technique combined



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 9

with chemometric analysis and image processing was a rapid, timesaving strategy for confirmation of content uniformity of pharmaceutical materials during in-process control.

(25) Tatavarti, A. S.; Fahmy, R.; Wu, H.; Hussain, A. S.; Marnane, W.; Bensley, D.; Hollenbeck, G.; Hoag, S. W. AAPS PharmSciTech. 2005, 6, E91-E99.

AUTHOR INFORMATION

(26) Tewari J.; Dixit V.; Malik, K. Sensors and Actuators B: Chemical. 2010, 144, 104-111. (27) Xie, Y.; Song, Y.; Zhang, Y.; Zhao, B. Spectrochim Acta A Mol Biomol Spectrosc. 2010, 75, 1535-9. (28) Burns, D. A.; Ciurczak, E. W. Handbook of near-infrared analysis, 3rd ed.; Taylor & Francis Group: New York, 2007. (29) Kandpal, L. M.; Lohumi S.; Kim, M. S.; Kang, J.; Cho, B. K. Sensors and Actuators B: Chemical. 2016, 229, 534-544.

Corresponding Author * Email: [email protected]. Tel: +82-42-821-6715. Fax: +82-42823-6246. * Email: [email protected]. Tel: 617-914-7189. 317488-8535.

ACKNOWLEDGMENT Author would like to acknowledge Patheon Toronto, Canada and Process analytical technology, Analytical development, Biogen, Cambridge, MA for the tableting work.

REFERENCES (1) Hughes, J. P.; Rees, S.; Kalindjian, S. B.; Philpott, K. L. Br J Pharmacol. 2011, 162, 1239-1249. (2) Blanco, M.; Bautista, M.; Alcalà, M. Drug Design, AAPS PharmSciTech. 2008, 9, 1130-1135. (3) Cebeci-Maltaş, D.; McCann, R.; Wang, P.; Pinal, R.; Romañach, R.; Ben-Amotz, D. Drug Design, J Pharm Innov. 2014, 9, 1-4. (4) Alvarenga, L.; Ferreira, D.; Altekruseb, D.; Menezes, J.; Lochmann, D. J Pharm Biomed Anal. 2008, 48, 62-69. (5) Wrezel, P.; Chion, I.; Hussain, M. LCGC NORTH AMERICA. 2004, 22, 10. (6) Laviana, L.; Llorente, I.; Bayod, M.; Blanco, D. J Pharm Biomed Anal. 2003, 32, 167-174. (7) Laviana, L.; Fernández-Marí, F.; Bayod, M.; Blanco, D. J Pharm Biomed Anal. 2003, 31, 321-8. (8) Sulub, Y.; Konigsberger, M.; Cheney, J. J Pharm Biomed Anal. 31, 321-8. 2011, 55, 429-434. (9) Liew, C.; Karande, A.; Heng, P. Int J Pharm. 2010, 386, 138148. (10) Schaefer, C.; Clicq, D.; Lecomte, C.; Merschaert, A.; Norrant, E.; Fotiadu, F. Talanta. 2014, 120, 114-125. (11) Flåten, G.; Ferreira, A.; Bellamy, L.; Frake, P. J Pharm Innov. 2012, 7, 38-45. (12) Amigo, J.; Cruz, J.; Maspoch, S.; Coello, J.; Blanco, M. TrAC Trends in Analytical Chemistry. 2008, 27, 696-713. (13) Amigo, J. Anal Bioanal Chem. 2010, 398, 93-109. (14) Franch-Lage, F.; Amigo, J. M.; Skibsted, E.; Maspoch, S.; Coello, J. Int J Pharm. 2011, 411, 27-35. (15) Osorio, J.; Stuessy, G.; Kemeny, G.; Muzzio, F. Chemical Engineering Science. 2014, 108, 244-257. (16) Palou, A.; Cruz, J.; Blanco, M.; Tomas, J.; Rı´os, J.; Alcala, M. J Pharm Anal. 2012, 2, 90-97. (17) Amigo, J.; Ravn, C. Eur J Pharm Sci. 2009, 37, 76-82. (18) Keerthi, M. L.; Kiran, R. S.; Rao, V. U. M.; Sannapu, A.; Dutt, A. G.; Krishna, K. S. Int. J. Sci. Rev. Res.2014, 40, 214-221. (19) Gendrin, C.; Roggo, Y.; Collet, C. J Pharm Biomed Anal. 2008, 48, 533-533. (20) Jaumot, J.; Gargallo, R.; Juan, A.; Tauler, R. Chemometrics and Intelligent Laboratory Systems. 2005, 76, 101-110. (21) Ravn, C.; Skibsted, E.; Bro, R. J Pharm Biomed Anal. 2008, 48, 554-561. (22) Storme-Paris, I.; Clarot, I.; Esposito, S.; Chaumeil, J. C.; Nicolas, A.; Rieutord, A.; Chaminade, P. Eur J Pharm Biopharm. 2009, 72, 189-98. (23) Rinnan, A.; Berg, F.; Engelsen, S. B. TrAC Trends in Analytical Chemistry. 2009, 28, 1201-1222. (24) Martínez, L.; Peinado, A.; Liesum L. Int J Pharm. 2013, 451, 67-75.


Page 9 of 9


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 ACS Paragon Plus Environment

In-Process Control Assay of Pharmaceutical Microtablets Using

Recommend Documents