Low-Frequency Raman Mapping and Multivariate Image Analysis for Complex Drug Products

• FDA

This article reflects the views of the authors and should not be construed to represent the Food and Drug Administration’s views or policies.

Confocal Raman mapping provides a robust means of characterization that can non-destructively provide both chemical and structural information for a wide range of pharmaceutical dosage forms.1-3 A literature review of Raman mapping for pharmaceuticals shows the spectral ranges used in studies is primarily limited to Stokes signals that are 200 to 4000 cm-1 away from the excitation laser line. This region is accessible with most commercial instruments and allows extracting valuable fingerprint information about molecules such as aromatics, carbonates, sulphates, silicates, oxides and hydroxides within the 500-1500 cm-1 range, and hydrogen interactions with carbon, nitrogen and oxygen at around 3000 cm-1.4 By contrast, the low-frequency Raman region (<10 cm-1 to 200 cm-1) has not been readily available in the past, even though Raman in this frequency range provides access to lattice vibrations of molecular crystals and has the potential to more directly probe intermolecular interactions in solids such as pharmaceutical dosage forms. Specifically, low-frequency Raman spectroscopy provides an avenue to probe polymorph detection of pharmaceutical systems as well as drug identification and quantitation for crystalline materials, both of which are critical quality attributes that need to be assessed and monitored during or after manufacturing.5-6 The low frequency Raman region for measurements of lattice phonons in pharmaceuticals has become more accessible in recent years with advances in precise optical filters and narrow linewidth lasers. Therefore, low-frequency region investigations in pharmaceutical products has started to transition from the academic laboratory with customized laboratory set-ups to pharmaceutical labs and production lines.7-10

This article focuses on low-frequency Raman spectroscopic analysis using a commercially available low-frequency Raman laser/filter coupled with a Raman microscope to perform pharmaceutical mapping measurements. In addition, multivariant statistical analysis techniques are utilized to highlight the unique and wide range of information that can be obtained from low-frequency Raman mapping of pharmaceuticals.

Subscribe to our e-Newsletters
Stay up to date with the latest news, articles, and events. Plus, get special offers
from American Pharmaceutical Review – all delivered right to your inbox! Sign up now!

Here we provide three examples of using low-frequency Raman mapping for analysis and characterization of a wide range of pharmaceutical drug types. First, we demonstrate the capability of low frequency Raman mapping for determining the active pharmaceutical ingredient (API) distributions and crystal sizes in an over-the-counter (OTC) oral dosage form. The mapping analysis of the OTC product Excedrin® was performed on a tablet core. Using multivariate analysis of Excedrin®, three major components were observed in the low frequency spectral region, all of which were similar to the reference spectra of the three declared APIs in the drug product (i.e., caffeine, acetaminophen and aspirin). In the second example, low-frequency Raman mapping was utilized for the identification of polymorph conversions in transdermal drug delivery systems (TDDSs). A principal component analysis (PCA) of the Raman map revealed unintended polymorph formation within the TDDS. Finally, a topical cream (Zovirax®) was mapped via the low-frequency spectral region. The multivariate analysis of this Raman map demonstrated an additional capability of low-frequency Raman spectroscopy; probing crystal orientation. The spectral variance within the low-frequency Raman map was used to identify various crystal orientations. Orientation identifications can be useful, as spectral variances in crystalline APIs can be misinterpreted as polymorphs. Overall, the potential of low frequency Raman mapping combined with multivariate statistical analysis is demonstrated here through analysis of a variety of drug products. We show that low-frequency mapping is a complementary tool to fingerprint region Raman mapping for rapid and sensitive identification of APIs, polymorphs and other crystal information that is obtained from lattice phonons only.

Materials and Methods

Pharmaceutical products were characterized by StreamHRTM Raman imaging using a Renishaw InVia confocal Raman microscope coupled with a TR-Micro Thz-Raman System from Ondax Inc. which provided the filter/laser combo needed to access the low-frequency region down to ~5 cm-1 in both stokes and anti-stokes regions. An 808 nm laser excitation source from the Ondax system, -70°C thermoelectrically cooled charge-coupled device (CCD) camera and a Leica DM2500 microscope were used for the Raman acquisition. A 1200/mm grating was used to disperse the light through a spectral range of -400 cm-1 to 900 cm-1. The objective and step size used for the low-frequency Raman mapping varied to best fi t the product being analyzed, but in general an optical montage was first obtained to allow for a representative region to be selected and mapped. Cosmic ray removal was done in WiRE 4.4 software using the width of feature and nearest-neighbor methods. Preprocessing and MCR/PCA analysis of the data was then carried out in Matlab R2017a using PLS Toolbox + MIA (Eigenvector Research, Inc., Version 8.6.2).

Sample preparation for the products selected for analysis were as follows: The Excedrin® tablet was prepared by splitting it in half with a razor blade and placing it on the microscope stage. The release liner of the transdermal delivery system (TDDS) was removed and the TDDS was placed backing side down on a gold coated microscope slide and placed on the microscope stage. The Zovirax® cream was thinly spread on a gold coated microscope slide and placed on the microscope stage.

Result and Discussion

API Mapping

To investigate the feasibility of low-frequency Raman mapping for characterization of pharmaceutical products, an over-the-counter (OTC) tablet consisting of three active pharmaceutical ingredients (APIs) was selected for analysis. This OTC product (Excedrin®), is a combination drug for the treatment of pain comprised of acetaminophen (250 mg), aspirin (250 mg), and caffeine (65 mg) as the active ingredients, all of which have well defined peaks in the low frequency Raman region.

The Ondax system with an 808 nm laser/filter set coupled with the Renishaw InVia Raman microscope was used to obtain the low frequency Raman mapping data from the Excedrin® tablet. The strength of the Raman signal in the low-frequency region from ~ 5 cm-1 to 200 cm-1 allowed for data to be acquired with the fastest (100 millisecond) acquisition times allowed by the Raman system. Raman signals in the low-frequency region generally have signal intensities of up to ~4.5 times those in the fingerprint region, allowing for faster mapping to be performed to obtain the same signal-to-noise ratios. The data set obtained on the Excedrin tablet was comprised of approximately 220k unique spectra taken across the entire tablet surface (Figure 1A). Traditional, spectra by spectra comparison across the 220k set of spectra is prohibitively time consuming. However, multivariant image analysis can be applied to rapidly interpret the Raman data. In multivariate methods, spectral features are utilized fully by revealing the variance within a spectral map and, simultaneously, dimensionally reducing the dataset. Typically, unsupervised methods such as PCA are used for chemical imaging and performs the task via eigen decomposition of the covariance of the spectral dataset. However, eigenvectors that are used in the dimensionality reduction are not true representatives of the Raman spectra, rather, they represent the contribution of a mixture of the Raman signatures (peaks) to the spectral variance. To allow for easily visualized data sets that can be quickly compared to a library of reference spectra, a better suited method is Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS). For the analysis of the Excedrin® tablet, MCR-ALS was applied to the Raman microscopy mapping with the assumption of a three-component system due to the 3 APIs expected in the system. Spectral pre-processing included standard normal variant (SNV) followed by Savitzky-Golay smoothing (0 order, 7 point) and then a baseline correction with a Whittaker filter (asymmetry=0.001, lambda=100). Since the data obtained is resulting from Raman mapping, the components can not only be extracted, but also visualized spatially to reveal the location of each individual component across the tablet surface (Figure 1B).

The MCR-ALS analysis neatly highlights three unique components which can be compared to reference library spectra (Figure 1C). When these MCR loadings are overlaid with expected spectra for the APIs that was obtained from the USP reference standards, all three components can be readily identified. With component 1 shown in red on the MCR scores image resulting from the acetaminophen and component 2 shown in green resulting from caffeine in very good agreement with the USP reference spectra for those two APIs. Component 3 shown in blue indicates a good agreement with aspirin, however an additional shoulder is observed at ~140 cm-1 that isn’t present in the reference spectra for aspirin. This additional peak is likely due to an excipient present in the Excedrin® tablet that isn’t included in the model that was built since the limitations were set to three components. These labeled excipients include benzoic acid, carnauba wax, FD&C blue No. 1, hydroxypropylcellulose, hypromellose, microcrystalline cellulose, mineral oil, polysorbate 20, povidone, propylene glycol, simethicone emulsion, sorbitan monolaurate, stearic acid and titanium dioxide. However, due to the expected amorphous nature of the majority of these excipients, they are unlikely to contribute well defined features in the low-frequency region, as opposed to the forest of spectral signatures that would be expected in the traditional fingerprint region. Thus, this helps to highlight the ability of low-frequency Raman mapping to provide API distributions across the tablet surface with minimal interference from the excipients present.

Polymorph Identification

In addition to rapidly identifying and mapping API location across the surface of a pharmaceutical product, low-frequency Raman is capable of identifying changes that may occur in the API due to polymorphism or co-crystallization which may negatively affect the product performance. The low-frequency Raman region is a sensitive screen for polymorph changes due to the impact of changes in the lattice vibrations on the low frequency spectra.

For example, a transdermal delivery system (TDDS) with visual crystal growth was investigated with low-frequency Raman mapping. Specifically, an area where two crystals were present side-by-side and visually had an observable contrast (Figure 2A). The two crystal area was mapped using an 100x objective with three second accumulations using three micron step sizes for a data set comprised of ~20k spectra. For this type of data set, PCA was applied due to the unknown nature arising from the observed crystals, in contrast to Excedrin® where MCR was expected to comprehensively resolve the three API system. Spectral pre-processing entailed standard normal variant (SNV) followed by Savitzky-Golay smoothing (0 order, 7 point) and mean centering. Principle component (PC) 1 explained 62.34% of the variance observed and from the scores image for PC1 (Figure 2B), could be attributed to the differences between the two crystals resulting in one having a high score for PC1 (very “red”) and one having a negative score for PC1 (very “blue”), with the surrounding TDDS matrix having a score close to zero. To represent this visually, the average spectra for the “red” crystal was compared with that of the “blue” crystal allowing for the key spectral differences to be observed. These spectra were then compared to the reference spectra for the expected API and did not show good agreement with the spectrum of either crystal that was formed in the TDDS (Figure 2C).

These preliminary results indicated that either the API had a polymorphic conversion when the API crystallized out of the TDDS formulation or that the API co-crystallized out with an excipient used to prepare the TDDS, such as the adhesive, causing the observed changes in the expected API spectra. Thus, low-frequency Raman was able to detect the difference in two crystals approximately 200 microns in size and less than 100 microns apart. To perform this using traditional approaches for polymorph identification such as X-ray powder diffraction or differential scanning calorimetry would not be feasible with the current state of the art for such instrumentation.

Effect of Crystal Orientation

The third system analyzed was Zovirax® cream which has a labeled acyclovir API concentration of 5%. Optical microscopy of this cream showed rectangular crystals ranging in size from 10 to 80 microns present amongst the white cream matrix. A low-frequency Raman map was acquired on two crystals present side-by-side using a 100x objective with 1-micron step sizes giving a data set of approximately ~1500 spectra (Figure 3A). Spectral pre-processing entailed SNV followed by Savitzky-Golay smoothing (0 order, 7 point) and then a baseline correction with a Whittaker filter (asymmetry=0.001, lambda=100). A two-component MCR model was used and this resolved two components that emphasized the differences between the crystals (red) and the background matrix comprising the white cream base (blue) (Figure 3B1). The MCR loadings are shown in Figure 3B2 where component 1 (blue) has a broad amorphous peak due to the excipient matrix while component 2 (red) has well defined features resulting from the crystals that are in good agreement with low-frequency Raman for acyclovir reference standard (yellow). When the MCR model was expanded to 4 components, the model showed the two crystals with two distinct spectra. The MCR scores image for components 1 (green) and 4 (blue) show two distinct crystals (Figure 3C1) with the primary differences in the MCR loadings being attributed to a peak at 37 cm-1 for the “blue” crystal and a shoulder at 90 cm-1 for the “green” crystal (Figure 3C2).

Upon initial investigation, these changes were attributed to a polymorphic difference between the two crystals, however further analysis of multiple crystals revealed that this difference was due to the orientation of the crystals. Because of the directional and polarized nature of the laser beam used for excitation and the rectangular shape of the acyclovir crystals, excitation was being obtained either along the horizontal or vertical axis resulting in distinct low-frequency Raman peaks from horizontal or vertical crystal orientation. While crystal orientation will not change the performance of the API, the knowledge of the impact of orientation is important to note to avoid false assumptions of presence of polymorph when none are present. Importantly, the orientation sensitivity can be avoided entirely if the laser beam is depolarized prior to being focused on the sample.

Conclusions

In summary, the capabilities of low-frequency Raman mapping in the characterization of APIs in a variety of pharmaceutical products was demonstrated in three case studies. By utilizing the strong low frequency Raman signatures that arise from crystalline API, rapid data acquisition and high selectivity of the low-frequency region for the API over excipients (which tend to be more amorphous in nature resulting in weaker low-frequency Raman signatures), identification, size and distribution determination of crystalline components in pharmaceutical products was established. Additionally, the low frequency spectral region was shown to be particularly sensitive to polymorphic changes in the API. When coupled with multivariant analysis techniques such as MCR and PCA, low-frequency Raman maps allowed changes in the API to be visualized across the product surface. With this technique, critical quality attributes, such as tablet homogeneity of API location and form, could be rapidly assessed for a wide range of drug products.

Acknowledgements

This project was supported, in part, by an appointment (H.Y.) to the Research Participation Program at the Center for Drug Evaluation and Research administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and the U.S. Food and Drug Administration.

We would also like to acknowledge Dr. Anjan Roy from Ondax, Inc. for his valuable insight in discussions related to this project and Dr. Ahmed Zidan from FDA/CDER/OPQ/OTR/ Division of Product Quality Research (DPQR) for his assistance with the transdermal studies.

References

1. Ewing, A. V.; Kazarian, S. G., Recent advances in the applications of vibrational spectroscopic imaging and mapping to pharmaceutical formulations. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 2018, 197, 10-29.
2. Gordon, K. C.; McGoverin, C. M., Raman mapping of pharmaceuticals. International Journal of Pharmaceutics 2011, 417 (1), 151-162.
3. Smith, G. P. S.; McGoverin, C. M.; Fraser, S. J.; Gordon, K. C., Raman imaging of drug delivery systems. Advanced Drug Delivery Reviews 2015, 89, 21-41.
4. Socrates, G., Infrared and Raman characteristic group frequencies: tables and charts. John Wiley & Sons: 2004.
5. Otaki, T.; Tanabe, Y.; Kojima, T.; Miura, M.; Ikeda, Y.; Koide, T.; Fukami, T., In situ monitoring of cocrystals in formulation development using low-frequency Raman spectroscopy. International Journal of Pharmaceutics 2018, 542 (1), 56-65.
6. Mah, P. T.; Fraser, S. J.; Reish, M. E.; Rades, T.; Gordon, K. C.; Strachan, C. J., Use of lowfrequency Raman spectroscopy and chemometrics for the quantification of crystallinity in amorphous griseofulvin tablets. Vibrational Spectroscopy 2015, 77, 10-16.
7. Hisada, H.; Inoue, M.; Koide, T.; Carriere, J.; Heyler, R.; Fukami, T., Direct high-resolution imaging of crystalline components in pharmaceutical dosage forms using low-frequency raman spectroscopy. Organic Process Research & Development 2015, 19 (11), 1796-1798.
8. Larkin, P. J.; Dabros, M.; Sarsfi eld, B.; Chan, E.; Carriere, J. T.; Smith, B. C., Polymorph characterization of active pharmaceutical ingredients (APIs) using low-frequency raman spectroscopy. Applied Spectroscopy 2014, 68 (7), 758-776.
9. Walker, G.; Römann, P.; Poller, B.; Löbmann, K.; Grohganz, H.; Rooney, J. S.; Huff , G. S.; Smith, G. P. S.; Rades, T.; Gordon, K. C.; Strachan, C. J.; Fraser-Miller, S. J., Probing pharmaceutical mixtures during milling: The potency of low-frequency raman spectroscopy in identifying disorder. Molecular Pharmaceutics 2017, 14 (12), 4675-4684.
10. Wang, H.; Boraey, M. A.; Williams, L.; Lechuga-Ballesteros, D.; Vehring, R., Low-frequency shift dispersive Raman spectroscopy for the analysis of respirable dosage forms. International Journal of Pharmaceutics 2014, 469 (1), 197-205.