Rui Chen

In this article, inline monitoring of active pharmaceutical ingredient (API) crystallinity by Fourier transform near-infrared (FT-NIR) during a hot melt extrusion process is described. The area of the peak at 6100 cm

 was used to track the transitions between different solid dispersions during a hot melt extrusion in real time. The NIR observations are further corroborated by an offline terahertz (THz) Raman imaging analysis. Overall, both high process temperature and high screw speed are conducive to the formation of amorphous acetaminophen (APAP) dispersions, but temperature appears to more directly impact the formation of the most desired amorphous dispersion. FT-NIR is effective in providing real-time information on the solid physical state transformation, but through bulk measurements. Raman microscopy complements FT-NIR by offering direct visualization of the spatial distributions of different solid physical states of the API.

Hot melt extrusion (HME) is gaining increased interest in the pharmaceutical industry because of its high manufacturing efficiency and economic benefits. HME has been utilized to develop versatile dosage forms, especially for poorly soluble active pharmaceutical ingredients (API) (1–3). In an HME process, APIs are compounded with thermoplastic polymers and extruded to form solid dispersion. The polymer matrix acts as a solid solvent for the drug molecules. Amorphous API dispersions in a polymer matrix at the molecular level is highly desirable because the drug is in a dissolved molecular state with no lattice energy to overcome prior to dissolution. Furthermore, amorphous API dispersions can provide a physical barrier to recrystallization or aggregation of drug particles and reduce the risk of physical change post-manufacture.

To ensure the consistency and the product quality, it is imperative to have a thorough understanding of the critical process parameters, such as screw design, rotational speed, and zone temperature, in real time as well as their impact on the quality of the final products. To that end, Fourier transform near-infrared (FT-NIR) spectroscopy in conjunction with multivariate analysis (MVA) techniques have long been used as an in-process analytical tool to provide molecular level chemical information through direct measurements of the material being processed (4–6). Recently, Repka and others reported the successful implementation of FT-NIR for inline API concentration and process stability monitoring during HME processes (7,8).

Another integral part of the overall HME process design and optimization is characterizing the solid physical state of extrudates (amorphous, crystalline, or partially amorphous), which provides critical information on product stability during shelf life, crystallization propensity, drug dissolution, and drug bioavailability, as well as process efficiency (2). Previously, we reported a method employing terahertz (THz) Raman imaging to characterize the solid state of the acetaminophen (APAP) within a solid dispersion prepared by HME (9). The low-frequency Raman shift region of ≤ 6 THz or ≤ 200 cm

, is sensitive to the variations in crystal structures, such as lattice form and orientation. Therefore, THz Raman spectroscopy readily identifies different polymorphs of APIs and monitors API crystallization and dissolution processes (10–12).

In this article, we describe the use of inline FT-NIR to characterize the solid physical states transformation of acetaminophen (APAP) in hydroxypropyl methylcellulose (HPMC) matrix prepared by HME. By tracking the changes in the peak area associated with crystalline APAP during HME, the underlying transitions between different solid dispersions under different process conditions are rationalized. The FT-NIR data is further corroborated by an offline THz Raman imaging analysis.

Acetaminophen (APAP) (Spectrum Chemicals), a crystalline biopharmaceutics classification system (BCS) I drug with a melting point of 169°–170 °C, was chosen as the model API. Hydroxypropyl methylcellulose (HPMC) (Ashland) E5 grade, a common hydrophilic matrix material for amorphous solid dispersions, was used as the polymer matrix.

An 11-mm twin-screw extruder, Thermo Scientific Process 11 was used for the HME processes. Figure 1 is a schematic showing the screw profile design as well as the process configuration. APAP and HPMC were tumbling mixed for 30 min before fed into the extruder at the Zone 1 position. The Zones 2 through 8 were mixing, conveying and discharge zones. The Zone 9 is a die that reduces the barrel configuration to a single circular orifice with 5 mm in diameter.

FIGURE 1: Schematic of a Process 11 twin screw extruder configured in this study.

The blending ratio of APAP and HPMC was 30:70 (w/w). Four different HME processes were performed with two different screw speeds at two different temperatures, as shown in Table I. For each HME process, after the extrusion reached the steady state, the extrudates from the die (zone 9) was collected and pressed into thin films using a hydraulic press with heating panels at 40 °C to be analyzed by Raman imaging.

A Thermo Scientific Antaris II MDS FT-NIR spectrometer equipped with a diffuse reflectance probe was used for spectral acquisition. The probe and a metal reflector were screwed opposite into the sensor ports of the die (zone 9) with a gap of 4 mm. NIR spectra were collected every 20 s with the spectral range of 4000–10,000 cm

 spectral resolution, and 16 scans were co-added (8 s scan time). FT-NIR data acquisition was performed using the Thermo Scientific Result v3 software package. Data processing and modeling was performed using the Thermo Scientific TQ Analyst software.

To track the polymorphous change of APAP during the extrusion processes, the spectral peak at 6100 cm

 was chosen as the indicator for crystalline APAP. All spectral data was first processed with 11-data point and third-order Savitzky–Golay smoothing. The peak area was then integrated between 6055.39–6124.81 cm

Raman spectroscopy and Raman imaging were performed using a Thermo Scientific DXR3xi Raman imaging microscope, using a 532 nm laser operated at 8 mW power at the sample. Data acquisition and processing were carried out using the Thermo Scientific OMNICxi software.

Raman images were obtained by the Correlation Profiling option of the OMNICxi software, using the THz region (≤ 200 cm

) of the crystalline APAP Raman spectrum as a reference. The correlation profile shows the correlation between the spectrum at each pixel (sampling point) in the Raman image and the specified reference spectrum. Each color in the rainbow-colored intensity scale represents a degree of correlation with the reference spectrum. The higher the intensity value, the greater similarity to the reference spectrum. A value of 1.0 indicates that the sample spectrum and reference spectrum are identical and is displayed as red. For the purpose of direct comparison, a correlation range of 0.75–1 were applied to all the Raman images.

As shown in Figure 2, all FT-NIR spectra have strong features in the first overtone C–H stretching region around 6000 cm

. The amorphous and crystalline APAP, however, are noticeably different (Figure 2). The crystalline APAP spectrum has a sharp peak around 6100 cm

, whereas the amorphous APAP has a broad feature around 6600 cm

FIGURE 2: (top) FT-NIR spectra acquired during the extrusion processes. (bottom) FT-NIR spectra of amorphous and crystalline APAP.

Spectral peak wavelengths are predominately determined by molecular structures, interactions among molecules, as well as their environment, and peak intensities are often proportional to component concentrations. Therefore, the area of crystalline APAP peak at 6100 cm

 is an indicator for the concentrations of crystalline APAP in the samples. When the extrusion process was operated at 130 °C and 50 rpm (PRC 1), the spectral peak at 6100 cm

 is apparent. As the extrusion condition moved to a higher screw speed (PRC 2), a higher extrusion temperature (PRC 3), or a combination of both (PRC 4), the crystalline peak at 6100 cm

 decreased while the amorphous peak at 6600 cm

 increased concurrently, indicating a solid physical state transformation from crystalline to amorphous. A high temperature promotes amorphization as well as diffusion of the API molecules through the polymer chains for adequate mixing. On the other hand, increasing the screw speed increases the dissolution of the drug into a molten carrier and dispersion by imparting the energy generated by the friction between screws and extruded material.

Figure 3a shows the calculated peak areas of crystalline APAP under different HME process conditions. Under PRC 1 (130 °C, 50 rpm), the extrudates exhibit the highest crystallinity. When the screw speed was increased to 200 rpm while maintaining the same temperature at 130 °C (PRC 2), the crystallinity decreased. When the process conditions were changed back to PRC 1, the crystallinity also reverted to its original value, suggesting the reversible nature of the transition from PRC 1 to PRC 2 and back to PRC 1. Next, the temperature was increased to 170 °C while maintaining the screw speed at 50 rpm (PRC 3). The crystallinity decreased again, but to a level below the value observed in PRC 2, as indicated by the gap bracketed by the two green arrows in Figure 3a. Clearly, PRC 3 resulted in an overall more amorphous solid dispersion than PRC 2. However, increasing the screw speed to 200 rpm (PRC 4) did not further decrease the crystallinity. Finally, the process was changed back to PRC 1. Interestingly, the crystallinity remained at the low value and did not change to the value previously observed in PRC 1.

FIGURE 3: (a) The top shows screw speed vs. time; the middle of the figure shows process temperature vs. time; and the bottom of the figure shows the area of the peak at 6100 cm

 vs. time. (b) Illustration of the three different types of solid dispersion.

A two-component, API-polymer carrier solid dispersion can form multiple structures depending on their composition and process- ing history (13), as illustrated in Figure 3b. The drug can be molecularly dispersed within the polymer matrix, forming a thermodynamically stable, homogeneous solid solution (SS), the most desirable structure of solid dispersions. Another form is a dispersion of crystalline drug particles in a polymer matrix, referred to as a crystalline solid dispersion (CSD). Alternatively, an intermediate, meta-stable structure where amorphous drug aggregates are dispersed in a polymer matrix (ASD) can be formed. The initial process changes from PRC 1 to PRC 2 and back to PRC 1 likely induced a transition of CSD to ASD and back to CSD, at least in portions of the sample, causing the rise and fall of the overall crystallinity value. Because ASD is a meta-stable state, it can be converted to the equilibrium CSD provided that the thermodynamic driving force is sufficient to overcome the Gibbs free energy barrier. The later transitions of PRC 1 to PRC 3 by raising the temperature to 170 °C resulted in the formation of SS, indicated by the lowest crystallinity value. Once the solid dispersion reached the SS, the polymer matrix provided both a physical barrier to aggregation and stabilization through the drug–polymer interaction. As the result, the API stays in that solid solution. Further transitions from PRC 3 to PRC 4 then back to PRC 1 did not change the crystallinity.

The observations in the FT-NIR experiments were further corroborated by THz Raman imaging analysis. At 130 °C and 50 rpm (PRC 1, Figure 4a), agglomeration of crystalline APAP (CSD, red regions), partially amorphous dispersion (ASD, green regions), and amorphous dispersion (SS, blue regions) coexist. Note that the processing temperature was well below the temperature of APAP (169 °C), and the energy imparted to APAP by screw rotation was not enough to accomplish complete amorphization. At this low processing temperature, even with a higher screw speed of 200 rpm (PRC 2, Figure 4b), the sample still contains noticeable amount of partially amorphous dispersion (ASD). Figures 4a and 4b are in sharp contrast to those of the extrudates processed at a higher temperature of 170 °C (PRC 3 and PRC 4; Figures 4c and 4d). By raising the temperature, near-complete amorphous APAP dispersion (SS) were formed with both screw speeds. The comparison between the four processes clearly illustrate that screw speed is a critical process variable that dictates the energy supply to the extruded material, the degree of mixing, and, to a lesser extent, the residence time. Increasing screw speed increases dissolution of drug into a molten carrier and dispersion. However, the higher temperature is more conducive to the formation of SS.

FIGURE 4: Raman images of extrudates from different HME processes. Images (a–c) were collected using a 50× objective and a 5 μm pixel size. Image (d) was collected using a 100× objective and a 0.5 μm pixel size.

In this study, inline monitoring of API crystallinity by FT-NIR during an HME process was demonstrated. The crystallinity of APAP, measured by the peak area at 6100 cm

, shows three distinct values, which are speculated to be associated with three different forms of solid dispersion: crystalline solid dispersion, amorphous solid dispersion, and solid solution. Overall, high process temperature and high screw speed are conducive to the formation of amorphous APAP dispersions, but temperature appears to have a more direct impact on the formation of the desired amorphous APAP dispersion. These observations are further corroborated by an offline THz Raman imaging analysis. The extrudates with the highest crystallinity value (least amorphous) show the coexistence of appreciable amount of CSD with ASD and SS, whereas the extrudates with the lowest crystallinity value (most amorphous) contain predominantly SS. FT-NIR is effective in providing real-time information on the solid physical state transformation, but through bulk measurements. Raman microscopy complements FT-NIR by offering direct visualization of the distribution of different solid physical states of the API.

(1) M.A. Repka, S. Bandari, V.R. Kallakunta, A.Q. Vo, H. McFall, M.B. Pimparade, and A.M. Bhagurkar, 

(2) S.D. Schaber, D.I. Gerogiorgis, R. Ramachandran, J.M.B. Evans, P.I. Barton, and B.L. Trout, 

(3) R. Censi, M.R. Gigliobianco, Cr. Casadidio, and P. Di Martino, 

(4) N. Furuyama, S. Hasegawa, T. Hamaura, S. Yada, H. Nakagami, E. Yonemochi, and K. Terada, 

(6) A. Paudel, D. Raijada, and J. Rantanen, 

(7) A.Q. Vo, H. He, J. Zhang, S. Martin, R. Chen, M.A. Repka, 

(8) A.Q. Vo, G. Kutz, H. He, S. Narala, S. Bandari, and M.A. Repka, 

(9) M. Ibrahim, J. Zhang, M.A. Repka, and R. Chen, 

(11) A. Paudel, D. Raijada, and J. Rantanen, 

(12) F. Tres, K. Treacher, J. Booth, L.P. Hughes, S.A.C. Wren, J.W. Aylott, and J. C. Burley, 

 are with Thermo Fisher Scientific. Direct correspondence to: 

Articles

Inline FT-NIR and offline terahertz Raman imaging analysis are used to characterize active pharmaceutical ingredient (API) crystallinity and to monitor different solid physical states of the API, to control process parameters of hot melt extrusion.

Solid Physical State Transformation in Hot Melt Extrusion Revealed by Inline Near-Infrared (NIR) Spectroscopy and Offline Terahertz (THz) Raman Imaging

In this study, several strains of yeast cells, both wild-type (WT) and engineered, were analyzed by Raman imaging microscopy. Two biologically active compounds produced within the engineered yeast cells were readily identified by their characteristic Raman spectral peaks. The locales and the relative abundance of those biologically active compounds were visualized from the Raman microscopic images of the individual cells. Raman imaging microscopy further reveals the differences in efficiencies with which the biologically active compounds were produced by different yeast cell strains.

In recent years, bioengineering of yeast cells has made them a “biological cell factory” for producing a range of chemicals, including biologically active compounds through fermentation processes. This “biological cell factory” can efficiently convert starch, sucrose, or lignocellulose-based raw materials into precursor metabolites, which can further be converted into the products of interest (1). Bulk chemical analyses, such as gas chromatography–mass spectrometry (GC–MS) and nuclear magnetic resonance (NMR), can be used for screening fermentation cultures, assessing product yield, and delivering process optimization. These methods, however, are destructive, time-consuming, and insufficient, because they fail to provide specificity at the cellular level (2). Fluorescence microscopy-based imaging is capable of visualizing large or small molecules at the cellular level, but the success of this approach largely hinges on the photostability and brightness of fluorescent labels, their possible perturbations to the biological system, and the methods by which the labels are attached (3). Confocal Raman microscopy-based imaging has emerged as a great alternative (3,4). Raman spectroscopy provides chemical annotations to the constituents within the cell, and the high spatial resolution intrinsic to confocal microscopy provides cellular level information. In contrast to the fluorescence microscopy-based imaging approach, Raman imaging is nondestructive and requires minimal sample preparation.

In this study, wild-type (WT) and several strains of engineered yeast cells were analyzed by Raman imaging. Two biologically active compounds produced within the engineered yeast strains were readily identified. Furthermore, it was revealed that different strains produced the two compounds with different efficiencies. Raman images of the individual cells show the locales as well as the relative abundance of those compounds. The information extracted provides valuable information that helps optimize the fermentation process.

A 2 μL sample of the WT and each engineered yeast cell strain sample was transferred to an aluminum-coated, 12-positions slide, and then it was air dried prior to the Raman imaging experiments. The two biologically active reference compounds are denoted as “compound 1” and “compound 2”, respectively. Compound 1 is a terpene and compound 2 is a sterol. For the two reference compounds in solid powder form, a small amount of each was transferred to a glass slide, and the Raman spectra were collected.

The Raman spectra and the Raman area images (for an area containing one or two cells within each sample) were acquired using a Thermo Scientific DXR3xi Raman Imaging Microscope. The key experimental parameters for Raman imaging are listed below:

Raman images were obtained by the correlation profiling option of the OM-NICxi software, using a representative Raman spectrum from within a cell as the reference. The correlation profile shows the correlation between the spectrum at each pixel (sampling point) in the Raman image and the specified reference spectrum. Each color in the rainbow-colored intensity scale represents a degree of correlation with the reference spectrum. The higher the intensity value, the greater similarity to the reference spectrum.

Figure 1 shows the representative Raman spectra of the two biologically active compounds, compound 1 (pink trace, top) and compound 2 (red trace, bottom). Both compounds have prominent features in the C-H stretching region (~2900 cm

). The two strong peaks attributed to the C=C stretch- ing vibrations are located at 1667 cm

, respectively, for compound 1 and compound 2. The downshift of the C=C stretching peak in compound 2 (1605 cm

) is because of a ring structure. There are no strong interfering Raman peaks from the cellular matrices in this frequency range. Hence, these peaks are chosen as the Raman markers to distinguish the two biologically active compounds in the yeast cell samples by Raman imaging.

FIGURE 1: Raman spectra of the pure biologically active compounds, (a) compound 1, and (b) compound 2. These two compounds can be clearly distinguished by the presence of strong C=C peaks at 1667 cm

Figure 2 shows the Raman image of a WT yeast cell (Figure 2a), the representative Raman spectra from the different locations of the imaged region (Figure 2b), and the representative Raman spectra of compound 1 and compound 2. As can be seen in Figure 2a, the cells (in red) can be spectroscopically differentiated from the surrounding media (in blue), with a clearly delineated contour in green color. As expected, the cells are devoid of either compound 1 or compound 2 (Figure 2c). Raman image of the strain B yeast cells is shown in Figure 3. Both compound 1 and compound 2 are produced in the cell, as evidenced by Figures 3b and 3c. Interestingly, for the same strain B, there are cell-to-cell variations in the level of produced compounds. Based on the relative peak intensities, the cell in the upper- left corner appears to produce more compound 1 than compound 2, whereas the cell in the center produces similar amounts of compound 1 and compound 2.

FIGURE 2: Raman image of a wild-type (WT) yeast cell. (a) Raman image of a selected region for the WT yeast cells. The Raman image was generated by correlation profiling. (b) Raman spectra from the spots S-1 and S-2 on the Raman image. (c) Raman spectra of S-1 and S-2, as well as compounds 1 and 2 displayed in the 1540–1750 cm

 region. The Raman spectra for the cells were normalized with respect to the C-H stretching peaks near 3000 cm

FIGURE 3: Raman image of the strain B yeast cells. (a) Raman image of a selected region for the strain B yeast cells. The Raman image was generated by correlation profiling. (b) Raman spectra from the spots S-1, S-2 and S-3 on the Raman image. (c) Raman spectra of S-1, S-2 and S-3, as well as compounds 1 and 2 displayed in the 1540–1750 cm

 region. The Raman spectra for the cells are normalized with respect to the C-H stretching peaks near 3000 cm

Figure 4 summarizes the Raman imaging analyses of the three different strains of yeast cells—strain A, strain C, and strain D—in terms of their yield in producing compound 1 and compound 2. At a quick glance, strain A appears to be a low producer of compound 1 and does not produce compound 2. Strains C and D, on the other hand, appear to produce both compound 1 and compound 2, but strain D has a higher yield for compound 1 than for compound 2.

FIGURE 4: Raman images of three different strains of yeast cells. (Top row) Raman images from the selected regions for the three different strains: strain A, strain C, and strain D. The Raman images were generated using correlation profiling. (Bottom row) Raman spectra (1540–1750 cm

 region) from the spots on the top corresponding Raman images. The Raman spectra are normalized with respect to the C-H stretching peaks near 3000 cm

Figure 5 summarizes the yield with which compound 1 and compound 2 are produced by the four different engineered strains of yeast cells (strains A–D). Strain A (blue) appears to be a low producer of compound 1 and does not produce compound 2. Strains B, C, and D all produce both compound 1 and compound 2. Strain B displays nonuniform levels of produced compounds from cell-to-cell, some cells (pink) producing higher level of compound 1 than the others (cyan). Strain C (green) seems to produce higher level of compound 2, and strain D (red) has the highest yield for compound 1. Caution should be exercised when assessing the product yields from the Raman peak intensity. A more rigorous, quantitative analysis requires the respective Raman cross sections for the compounds 1 and 2 be properly accounted for.

FIGURE 5: Raman spectra of the strains A–D of the yeast cells. (a) Raman spectra (1540–1750 cm

 region) from the cells of the strains A–D, shown in the offset mode, and (b) in the overlayed mode. The Raman spectra are normalized with respect to the C-H stretching peaks near 3000 cm

In this study, Raman imaging was successfully applied to analyze different strains of yeast cells, both WT and engineered. The target bioactive compounds produced by the engineered yeast cells can be readily identified and differentiated by their respective Raman spectra, allowing for assessment of the product yields of different yeast strains by Raman imaging. Raman images of the individual cells from the four engineered yeast cell strains indicate that these strains produce compounds 1 and 2 with different efficiencies. Raman imaging can also be used to estimate the relative yield of the bioactive compounds. A quantitative analysis of product yield, however, requires a more systematic study on the Raman cross-sections of those compounds. The information obtained from the Raman imaging will be beneficial for a deeper understanding of the fermentation process and for optimization of the process for higher product yield.

The authors would like to acknowledge Dr. Michael Leavell and his colleagues at Amyris for providing the samples and many insightful discussions.

(1) Y. Chen, L. Daviet, M. Schalk, V. Siewers, and J. Nielsen, 

(2) R. Miyaoka, M. Hosokawa, M. Ando, T. Mori, H.O. Hamaguchi, and H. Takeyama, 

(4) L. Mikoliunaite, R.D. Rodriguez, E. Sheremet, V. Kolchuzhin, J. Mehner, A. Ramanavicius, and D.R.T. Zahn, 

 are with Thermo Fisher Scientific in Madison, Wisconsin. 

 was formerly a scientist with Amyris in Emeryville, California, when this work was being conducted. Direct correspondence to: 

Using Raman imaging, wild-type and engineered yeast cells were compared for their ability to produce bioactive compounds. Raman imaging microscopy is able to visualize locales, relative abundance, and production efficiencies of biologically active compounds for the individual yeast cells.

Rui Chen

Articles

Solid Physical State Transformation in Hot Melt Extrusion Revealed by Inline Near-Infrared (NIR) Spectroscopy and Offline Terahertz (THz) Raman Imaging

Tracking Bioactive Compounds Produced by Genetically Engineered Yeast Cells Using Raman Imaging