Near-Infrared Spectroscopy for Cocrystal Screening. A Comparative

Anal. Chem. 2008, 80, 7755–7764

Near-Infrared Spectroscopy for Cocrystal Screening. A Comparative Study with Raman Spectroscopy Morten Allesø,† Sitaram Velaga,‡ Amjad Alhalaweh,‡ Claus Cornett,† Morten A. Rasmussen,§ Frans van den Berg,§ Heidi Lopez de Diego,| and Jukka Rantanen*,† Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, and Department of Food Science, Faculty of Life Sciences, University of Copenhagen, Copenhagen, Denmark, Department of Health Science, Luleå University of Technology, Luleå, Sweden, and Analytical R&D, H. Lundbeck A/S, Valby, Denmark Near-infrared (NIR) spectroscopy is a well-established technique for solid-state analysis, providing fast, noninvasive measurements. The use of NIR spectroscopy for polymorph screening and the associated advantages have recently been demonstrated. The objective of this work was to evaluate the analytical potential of NIR spectroscopy for cocrystal screening using Raman spectroscopy as a comparative method. Indomethacin was used as the parent molecule, while saccharin and L-aspartic acid were chosen as guest molecules. Molar ratios of 1:1 for each system were subjected to two types of preparative methods. In the case of saccharin, liquid-assisted cogrinding as well as cocrystallization from solution resulted in a stable 1:1 cocrystalline phase termed IND-SAC cocrystal. For L-aspartic acid, the solution-based method resulted in a polymorphic transition of indomethacin into the metastable r form retained in a physical mixture with the guest molecule, while liquid-assisted cogrinding did not induce any changes in the crystal lattice. The good chemical peak selectivity of Raman spectroscopy allowed a straightforward interpretation of sample data by analyzing peak positions and comparing to those of pure references. In addition, Raman spectroscopy provided additional information on the crystal structure of the INDSAC cocrystal. The broad spectral line shapes of NIR spectra make visual interpretation of the spectra difficult, and consequently, multivariate modeling by principal component analysis (PCA) was applied. Successful use of NIR/PCA was possible only through the inclusion of a set of reference mixtures of parent and guest molecules representing possible solid-state outcomes from the cocrystal screening. The practical hurdle related to the need for reference mixtures seems to restrict the applicability of NIR spectroscopy in cocrystal screening. * To whom correspondence should be addressed. E-mail: [email protected]. Telephone: +45 35336000. Fax: +45 35336030. Address: Department of Pharmaceutics and Analytical Chemistry, Faculty of Pharmaceutical Sciences, University of Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark. † Department of Pharmaceutics and Analytical Chemistry, University of Copenhagen. ‡ Luleå University of Technology. § Department of Food Science, University of Copenhagen. | H. Lundbeck A/S. 10.1021/ac8011329 CCC: $40.75  2008 American Chemical Society Published on Web 09/18/2008

Interest in the design and preparation of pharmaceutical cocrystals is increasing in the field of drug development. A cocrystal is a multicomponent system composed of the active pharmaceutical ingredient (API; neutral or ionized) and one or more guest molecules. In contrast to solvates, the guest molecule of a cocrystal is solid under ambient conditions.1 Cocrystals are stabilized through a variety of different intermolecular interactions including hydrogen bonds, aromatic π-stacking, and van der Waals forces, and unlike for salt forms, no proton transfer occurs between API and the guest molecule. Design of cocrystals aims at controlling drug and manufacturing-related properties such as powder flowability, powder compactability, and, in particular, dissolution rate.2 As for API polymorphs, intellectual property rights can be acquired for API cocrystals, thus providing unique safeguarding of the drug candidate.3 It is therefore of utmost importance that a thorough and reliable screening for potential cocrystal candidates be performed at an early stage of preformulation. Bis and co-workers4 recently carried out a comprehensive screening study of multiple API and guest molecule combinations, containing permutations of OH, N (aromatic), and CN moieties. The study demonstrated how an array of potential guest molecules is tested against one or more APIs via different methods of cocrystallization such as solvent evaporation, liquid-assisted cogrinding, and solid-state cogrinding. The solid product of each reaction chamber is commonly analyzed using X-ray powder diffractometry (XRPD) as the primary technique. XRPD detects changes in the crystal lattice and is therefore a powerful tool for studying polymorphism, pharmaceutical salts, and cocrystalline phases. XRPD can also be used as a part of automated robotic systems in high-throughput screening technologies.5,6 Drawbacks of this technique are long measurement times, insensitivity toward isostructural crystals, and, for some materials, preferred orientation effects. A time-consuming analysis may, particularly for high(1) Shan, N.; Zaworotko, M. J. Drug Discov.Today 2008, 13, 440–46. (2) Vishweshwar, P.; McMahon, J. A.; Bis, J. A.; Zaworotko, M. J. J. Pharm. Sci. 2006, 95, 499–516. (3) Trask, A. V. Mol.Pharmaceutics 2007, 4, 301–09. (4) Bis, J. A.; Vishweshwar, P.; Weyna, D.; Zaworotko, M. J. Mol.Pharmaceutics 2007, 4, 401–16. (5) Park, A.; Chyall, L. J.; Dunlap, J.; Schertz, C.; Jonaitis, D.; Stahly, B. C.; Bates, S.; Shipplett, R.; Childs, S. Expert Opin.Drug Discov. 2007, 2, 145– 54. (6) Storey, R.; Docherty, R.; Higginson, P.; Dallman, C.; Gilmore, C.; Barr, G.; Dong, W. Crystallogr.Rev. 2004, 10, 45–56.

Analytical Chemistry, Vol. 80, No. 20, October 15, 2008

7755

Figure 1. Molecular structures. The active pharmaceutical ingredient (API), indomethacin and guest molecules saccharin (SAC) and L-aspartic acid (AAC).

throughput screenings, create a bottleneck and thus reduce screening efficiency. Consequently, many researchers have turned to Raman spectroscopy as an alternative to XRPD.7-9 Raman spectroscopy probes the effect of crystal structure on bond vibrational energies and is potentially able to selectively distinguish between polymorphs of a given API. Furthermore, measurements are noninvasive, nondestructive, and rapid (seconds rather than minutes), which makes Raman spectroscopy ideal for automated high-throughput systems, for instance. The analytical potential of this technique in cocrystal screening has been acknowledged, though in-depth documentation on its use in this field is still sparse.10 A drawback of Raman spectroscopy is linked to fluorescence, which can markedly obscure the chemical features of the obtained spectrum. Compounds that exhibit high-intensity fluorescence are therefore less suitable for Raman analysis,11 though switching the laser wavelength sometimes solves this issue.12 The vibrational technique near-infrared (NIR) spectroscopy offers sampling-related advantages similar to Raman, but has no problem with fluorescence. Also, NIR instruments with integrated fiber-optic technology are markedly less expensive than the comparable Raman instruments. The price to pay for this convenience is a much less interpretable spectral signature in NIR spectroscopy. Since Raman and NIR spectroscopy are complementary at the molecular level (which is particularly relevant to the probing of water), using the two techniques together often provides increased understanding of solid-state phenomena.13-16 Aaltonen et al. recently demonstrated how NIR spectroscopy in (7) Morissette, S. L.; Soukasene, S.; Levinson, D.; Cima, M. J.; Almarsson, Ö. Proc. Natl. Acad. Sci. U.S.A 2003, 100, 2180–84. (8) Peterson, M. L.; Morissette, S. L.; McNulty, C.; Goldsweig, A.; Shaw, P.; LeQuesne, M.; Monagle, J.; Encina, N.; Marchionna, J.; Johnson, A.; Gonzalez-Zugasti, J.; Lemmo, A. V.; Ellis, S. J.; Cima, M. J.; Almarsson, Ö. J. Am. Chem. Soc. 2002, 124, 10958–59. (9) Hilfiker, R.; Berghausen, J.; Blatter, F.; Burkhard, A.; De Paul, S. M.; Freiermuth, B.; Geoffroy, A.; Hofmeier, U.; Marcolli, C.; Siebenhaar, B.; Szelagiewicz, M.; Vit, A.; von Raumer, M. J.Therm.Anal.Calorim. 2003, 73, 429–40. (10) Childs, S. L.; Hardcastle, K. I. Cryst.Eng.Comm. 2007, 9, 363–66. (11) Desrosiers, P. J. Mod.Drug Discovery 2004, 40–43, January. (12) McCreery, R. L. Raman Spectroscopy for Chemical Analysis., John Wiley & Sons, Inc.: 2000. (13) Heinz, A.; Savolainen, M.; Rades, T.; Strachan, C Eur.J.Pharm.Sci 2007, 32, 182–92. (14) Jørgensen, A.; Rantanen, J.; Karjalainen, M.; Khriachtchev, L.; Räsänen, E.; Yliruusi, J. Pharm. Res. 2002, 19, 1285–91.

7756


combination with fiber optics can be used for rapid and reliable solid-state analysis in polymorph screening.17,18 In spite of the vast amount of literature published on NIR spectroscopy for solid form analysis and its apparent advantages within the field, little work has been done so far to establish the usefulness of the technique for cocrystal screening. Therefore, the applicability of NIR spectroscopy in this context merits closer examination. The aim of this study was to investigate the analytical potential of NIR spectroscopy as a qualitative tool for cocrystal screening. The study was based on indomethacin as parent molecule and saccharin and L-aspartic acid as guest molecules (Figure 1). Saccharin is claimed to form a stable cocrystal with indomethacin,19 while L-aspartic acid is not expected to form any cocrystalline phase. The solid material generated by a solution-based method and liquid-assisted cogrinding for these two model systems was measured by Raman and NIR spectroscopy, and the approaches taken to data interpretation for the two spectroscopic techniques are compared and discussed. EXPERIMENTAL SECTION Materials. Indomethacin (γ form), saccharin, L-aspartic acid (purity >99.9%), ethyl acetate, and methanol (purity >99.8%) were purchased from Sigma Aldrich. All chemicals and solvents were used as received. MilliQ water was used throughout the study. For NMR, deuterated DMSO (d6, > 99.8% deuteration) and trifluoroacetic acid were purchased from Sigma Aldrich. Sample Preparation. (1) Solution-Based Method. A 1:1 mixture of indomethacin (72 mg, 0.20 mmol) and saccharin (37 mg, 0.20 mmol) was added to 25 mL of ethyl acetate in a 100-mL beaker and heated to aid dissolution. The solution was allowed to evaporate slowly in a fume hood to produce cocrystals. In a similar manner, indomethacin (72 mg, 0.20 mmol) and L-aspartic acid (27 mg, 0.20 mmol) were recrystallized from 60 mL of methanol/water (50% v/v) in a 100-mL beaker. (15) Vrecer, F.; Vrbinc, M.; Meden, A. Int. J. Pharm. 2003, 256, 3–15. (16) De Beer, T. R. M.; Allesø, M.; Goethals, F.; Coppens, A.; VanderHeyden, Y.; De Diego, H. L.; Rantanen, J.; Verpoort, F.; Vervaet, C.; Remon, J. P.; Baeyens, W. R. G. Anal. Chem. 2007, 79, 7992–8003. (17) Aaltonen, J.; Strachan, C. J.; Po¨lla¨nen, K.; Yliruusi, J.; Rantanen, J. J.Pharm.Biomed.Anal. 2007, 44, 477–83. (18) Aaltonen, J.; Rantanen, J.; Siiria¨, S.; Karjalainen, M.; Jørgensen, A.; Laitinen, N.; Savolainen, M.; Seitavuopio, P.; Louhi-Kultanen, M.; Yliruusi, J. Anal. Chem. 2003, 75, 5267–73. (19) Basavoju, S.; Bostro ¨m, D.; Velaga, S. P. Pharm. Res. 2007, 25, 530–41.

(2) Liquid-Assisted Cogrinding. A 1:1 mixture of indomethacin (143 mg, 0.40 mmol) and saccharin (73 mg, 0.40 mmol) was placed in a 10-mL Retsch grinding jar, and one drop of ethyl acetate was added. The mixture was ground for 15 min in a Retsch grinder (Mixer Mill MM301, Retsch GmbH & Co.). The powder was then dried and collected for further analysis. In a similar manner, indomethacin (143 mg, 0.40 mmol) and L-aspartic acid (53 mg, 0.40 mmol) were coground along with one drop of methanol/ water (50% v/v). (3) Preparation of Indomethacin Solid Forms. R-Indomethacin was obtained by dissolving γ-indomethacin in ethanol at 80 °C followed by antisolvent precipitation using water at 22 °C. The precipitated crystals were removed by filtration and then dried under vacuum at 36 °C. Amorphous indomethacin was prepared by melting γ-indomethacin in an aluminum pan on a laboratory hot plate at 165 °C for 3 min. The melt was slowly cooled to ambient temperature in a desiccator over phosphorus pentoxide and then lightly ground with a mortar and pestle. The solid-state purity of the resulting substances was verified with XRPD and differential scanning calorimetry (DSC). (4) Preparation of Physical Mixtures. Binary mixtures of saccharin and each of the three indomethacin solid forms (γ, R, and amorphous) were created in three molar ratios, 1:2, 1:1, and 2:1, by careful geometric mixing of the components. A total of 1 g per sample was prepared. A similar series of nine binary mixtures was created for the L-aspartic acid model system. All mixtures were measured by NIR spectroscopy. Instrumentation. (1) X-ray Powder Diffractometry. XRPD patterns of sample material obtained from the two preparative methods as well as the single components (i.e., indomethacin solid forms and guest molecules) were collected on a calibrated (using a silicon standard) Siemens DIFFRACplus 5000 powder diffractometer with Cu KR radiation (1.540 56 Å). The tube voltage and current were set to 40 kV and 40 mA, respectively. The divergence slit and antiscattering slit settings were variable for the illumination on the 20-mm sample size. The samples were filled into the cavity of a zero background sample holder or placed on a glass slide. Each sample was scanned between 5° and 50° in 2θ with a step size of 0.02°. The measurement time per step was 3.2 s. The scans were accumulated successively. Solid-state purity of γ- and R-indomethacin was verified by comparing experimental diffractograms to the calculated patterns obtained from the Cambridge Structural Database20 (reference codes: γ form)INDMET03; R form)INDMET02). (2) Differential Scanning Calorimetry. Thermal analyses of the generated indomethacin solid forms were performed on a Thermal Advantage DSC Q1000 V9.8 Build 296 (TA instruments-Waters, LLC) module, which was calibrated for temperature and cell constants using indium and sapphire. Samples (1-2 mg) were crimped in nonhermetic aluminum pans (30 µL) and scanned at a heating rate of 10 °C · min-1 in the range 25-300 °C under a continuously purged dry nitrogen atmosphere (flow rate 50 mL · min-1). The instrument was equipped with a refrigerated cooling system. The data were collected in triplicate for each sample and were analyzed using TA Instruments Universal Analysis 2000 V4.3A software. (20) Allen, F. H. Acta Crystallogr.B 2002, 58, 380–88.

(3) Nuclear Magnetic Resonance (NMR) Spectroscopy. 1H NMR was used to determine the molar ratio between indomethacin and L-aspartic acid in sample material prepared by liquid-assisted cogrinding and the solution-based method. Samples were dissolved by adding solvent (deuterated DMSO and additionally 200 µL of trifluoroacetic acid, the latter only for sample material prepared by the solution-based method) until no visible traces of solid material were present. Spectra were acquired using a Bruker (Rheinstetten, Germany) av 400 wb NMR operating at 400 MHz. In order to ensure reliable quantitation, 32 768 data points were acquired, zero-filled to 262 144 data points and processed with an exponential line broadening of 1 Hz prior to Fourier transformation resulting in 131 072 data points, the real part of the spectrum. Also, a 90° pulse, a relaxation delay of 30 s, and an acquisition time of 3.4 s, corresponding to a repetition time of ∼33.4 s, were used for all spectra, thus ensuring fully relaxed spectra. (4) Raman Spectroscopy. Raman spectra of sample material obtained from the two preparative methods as well as single components (i.e., indomethacin solid forms and guest molecules) were recorded. Approximately 10 mg of sample was transferred via a funnel to an integrated sample holder for a Perkin-Elmer near-IR FT-Raman 1700X spectrometer equipped with an indiumgallium-arsenide (InGaAs) detector. Spectra were collected using an excitation wavelength of 1064 nm of Nd:YAG laser radiation (power 600 mW). A total of 32 scans per spectrum was acquired in the range 4000-400 cm-1. Spectral resolution was 4 cm-1. Triplicate measurements were performed, each including a repeated sample transfer via a funnel. (5) Near-Infrared Spectroscopy. NIR spectra of sample material obtained from the two preparative methods, single components (i.e., indomethacin solid forms and guest molecules), and physical mixtures were recorded. Spectra were obtained in the reflectance mode using an FT-NIR spectrometer (Thermo Fisher Scientific, Nicolet Antaris Near-IR Analyzer) equipped with an InGaAs detector and a quartz halogen lamp. A total of 32 scans per spectrum was acquired in the range 10000-4000 cm-1. Spectral resolution was 8 cm-1. Prior to analysis, each sample was transferred to a glass vial having an inner diameter of 15 mm. Subsequently, samples were measured in random order through the bottom of the glass vial. Measurements were performed in triplicate at different random vial positions. Spectral Preprocessing. (1) Raman Spectroscopy. A thirdorder polynomial baseline correction was applied to all spectra using an in-house MATLAB routine (The MathWorks, Natick, MA, version 7.4/R2007a). This was done to remove fluorescent backgrounds and thereby make visual interpretation possible. (2) Near-Infrared Spectroscopy. NIR raw spectra suffered from a marked scatter effect observed as baseline shift and tilt as a result of varying particle size and varying shape of the sample material (the latter assessed visually from the powdered samples). In order to best preserve the original features of the NIR spectra and thus allow visual inspection, baseline correction was performed using standard normal variate (SNV) scaling as implemented in SIMCA-P+ 11.5 (Umetrics, Umeå, Sweden). Multivariate Visualization of NIR Data. Principal component analysis (PCA) was used to visualize the differences in the NIR Analytical Chemistry, Vol. 80, No. 20, October 15, 2008

7757

Figure 2. XRPD patterns of indomethacin/L-aspartic acid solid product obtained using the solution-based method and liquid-assisted cogrinding. Reference patterns of the pure components obtained from Cambridge Structural Database are also shown. To allow a visual comparison, the diffractogram of solid product from the solution-based method was scaled to an intensity of 300 across the entire 2θ range, while the coground sample and references were scaled to 100.

data. The following pretreatment procedures were performed prior to PCA. The NIR spectral region 8000-4100 cm-1 was used for analysis, excluding the third overtone region, in which sensitivity is lowest. Spectra were baseline corrected by SNV scaling, followed by a first derivative with 11-point Savitzky-Golay smoothing using SIMCA-P+ 11.5 (Umetrics). Last, spectra were mean centered. This preselection of spectral regions and subsequent preprocessing was carried out to enhance the accuracy and interpretation of the principal component models. PCA was calculated separately for the two model systems (i.e., saccharin- and L-aspartic acid- containing samples) using SIMCAP+ (Umetrics, version 11.5). The corresponding two score plots (one for each model system), which indicate the (dis)similarity between sample spectra, were constructed in MATLAB (The MathWorks, version 7.4/R2007a). RESULTS AND DISCUSSION Preliminary Sample Characterization. XRPD was used as a primary technique to characterize the solid material obtained from the saccharin (SAC) and L-aspartic acid (AAC) model systems. In the current study, γ-indomethacin and SAC subjected to evaporative crystallization and liquid-assisted cogrinding resulted in a 1:1 cocrystalline phase, henceforth abbreviated INDSAC cocrystal. This finding is in accordance with previously published XRPD data on IND-SAC cocrystal.19 Based on previous investigations performed by our group, the second model system, comprising γ-indomethacin and AAC, was expected not to form any cocrystalline phase. As can be seen from Figure 2, XRPD reflections from only the parent molecule (γ-indomethacin) and guest molecule (AAC) are present in the sample treated by liquid-assisted cogrinding. This clearly suggests a physical mixture and that indeed no cocrystal formation has taken place. NMR results established the molar ratio of the mixture as being 1:1 (Figure 3a). Surprisingly, sample material from the solution-based method shows a different XRPD pattern, though it still includes characteristic reflections from AAC. Since the guest molecule is still present in the sample, it seems likely 7758


that indomethacin has undergone a polymorphic transition induced by the recrystallization step. The new polymorph is identified as the metastable R form of indomethacin (Figure 2) and its formation is supported to some extent by another study performed under similar experimental conditions (albeit with no guest molecule present).21 Inspection of NMR data (Figure 3b) confirms a 1:0.7 molar ratio of indomethacin and AAC (∼41 mol % AAC), which is slightly lower than the anticipated 1:1 ratio. This discrepancy could be related to poor mixture homogeneity of R-indomethacin and AAC, which ultimately affects the sampling for NMR, a finding that was indeed later confirmed by Raman spectroscopy; see below. The results from the AAC system demonstrate the sometimes complex nature of the solid outcome produced from cocrystal screens. Even when no cocrystal is formed, the solid-state characteristics of the starting material can change, in this particular case due to a change in the crystal structure of a single component. Clearly, this should be kept in mind when analyzing the data. Raman Spectroscopy for Cocrystal Screening. (1) Indomethacin: SAC. Analysis of Raman data from a cocrystal screening is performed in a manner similar to that for XRPD, i.e., by searching for the absence and occurrence of peaks originally found in spectra of the single components. This is made possible by the high selectivity of Raman measurements. From Figure 4 it is readily apparent that solid material obtained by the solutionbased method and liquid-assisted cogrinding share similar solidstate properties. Moreover, it still contains spectral features of the two single components, γ-indomethacin and SAC, though with several peak shifts. In addition, no remnants of R-indomethacin and amorphous indomethacin are present. This suggests the formation of a cocrystalline phase and not a physical mixture of the two components. An advantage of Raman spectroscopy is the ability to selectively identify the molecular bonds involved in the stabilization of a given crystal. For the IND-SAC cocrystal, a hypsochromic and bathochromic shift to 1720 and 1684 cm-1, (21) Hancock, B. C.; Parks, M. Pharm. Res. 2000, 17, 397–404.

Figure 3. 1H NMR spectra of indomethacin/L-aspartic acid (IND:AAC) samples prepared using (a) liquid-assisted cogrinding and (b) the solutionbased method. The molar ratio between indomethacin and L-aspartic acid is determined by the integrals of the chemical shift of the methine groups (CH) located at approximately 6.6 (doublet) and 4.2 (double doublet) ppm for indomethacin and L-aspartic acid, respectively. The signal at 4.2 ppm for (b) cocrystallized IND:AAC is shifted and broadened due to solvent effects.

Figure 4. Baseline-corrected Raman spectra of IND-SAC cocrystal prepared using the two methods along with single-component spectra. The Y-axis has been scaled and offset to improve the clarity of the spectra.

respectively, is noted when comparing to the spectra of the two single components. Overall, these changes occur in the CdO stretch region of the vibrational spectrum. Due to the overlap of the CdO stretch (1699 cm-1) in pure γ-indomethacin and SAC, it is not possible to determine the nature of the two shifts observed in the cocrystal. However, previous studies suggest that one of the shifts is associated with the hydrogen bonded imide dimer of the IND-SAC cocrystal19 (Figure 5), while the other is related to the benzoyl group of indomethacin (originally at 1699 cm-1).22 In contrast to pure crystalline indomethacin, the benzoyl group is not hydrogen bonded in the IND-SAC cocrystal.19 Instead, the stabilization of the cocrystal is achieved via carboxylic acid dimers, as depicted in Figure 5.

Differences in other parts of the fingerprint region are also detected, for instance, a relative change in the intensity of two peaks at 1150-1200 cm-1 (compared to SAC). Interpretation in this region is complicated by overlapping group frequencies, though C-N and/or C-O stretch related to hydrogen bonding in the IND-SAC cocrystal are reasonable suggestions for band assignment. (2) Indomethacin: AAC. Evidently, the solid materials obtained using the two preparative techniques have different solid-state properties, which is supported by XRPD. The spectrum of solid material from the solution-based method resembles the spectrum (22) Taylor, L. S.; Zografi, G. Pharm. Res. 1997, 14, 1691–98.


7759

Figure 5. Hydrogen bonding involved in the stabilization of the IND-SAC cocrystal.

Figure 6. Baseline-corrected Raman spectra of indomethacin/L-aspartic acid solid product prepared using the two methods along with singlecomponent spectra and addition spectra for two selected combinations of API and guest molecule. Addition spectra are calculated based on the pure single-component spectra (i.e., γ/R-indomethacin and L-aspartic acid). The Y-axis has been scaled and offset to improve the clarity of the spectra.

of pure R-indomethacin whereas the γ form is evident in the spectrum from the liquid-assisted cogrinding experiment (Figure 6). Surprisingly, only a few traces of AAC are found in both spectra. For Raman spectroscopy in general, there may be two reasons for this. First, the two components may have markedly different Raman activities, causing the spectral features of the least Raman active constituent to be suppressed or completely absent from the mixture spectrum. Second, subsampling may be a cause, whichsfor physical mixturessis often observed at a combination of low sample homogeneity and low sampling volume of the Raman instrument, the latter being the result of small laser spot size and short penetration depth of the incident light.23 To address these issues, one would have to inspect addition spectra of relevant binary combinations of API solid form and guest molecule. This (23) Wikstro ¨m, H.; Lewis, I. R.; Taylor, L. S. Appl. Spectrosc. 2005, 59, 934–41.

7760


is feasible, as each peak contained in a Raman spectrum is associated with a specific vibration in the molecule, andsin the case of physical mixturesslinearity is therefore expected. The relevant combinations for this study are γ- and R-indomethacin, each in a 1:1 ratio with AAC. Evidently, there is a high degree of similarity between the liquid-assisted coground solid product and the addition spectrum of γ-indomethacin and AAC (Figure 6). The few traces of AAC in the experimental spectrum are therefore due to large differences in the Raman activities of the two constituents, with γ-indomethacin being markedly more Raman active than AAC. The same is not true, however, for the solutionbased method, as the addition spectrum contains clear peaks assigned to AAC. The absence of AAC peaks in the experimental spectrum is therefore not attributable to a difference in Raman activity. Rather, subsampling is a probable cause of this observa-

Figure 7. SNV-corrected NIR spectra of IND-SAC cocrystal prepared using the two methods along with single-component spectra. The Y-axis has been offset to improve the clarity of the spectra.

tion. This is especially an issue for the FT-Raman instrument used in the current work, where each sample is prepared in a fixed holder for subsequent analysis, while having an estimated spot size of only 600-700 µm in diameter. Similar conclusions were drawn in a study investigating the quantification of ternary mixtures of γ-, R-, and amorphous indomethacin by Raman spectroscopy, in which differences in particle density and the morphology (size/shape) of each indomethacin solid form compromised mixture homogeneity markedly in the case of stationary Raman measurements.13 Moreover, solid material from the solution-based method was not ground, which resulted in larger particles than those obtained with the liquid-assisted cogrinding method. Subsampling is thus the most reasonable explanation for the lack of clear AAC peaks in the experimental spectrum. As is evident for both model systems, the high chemical peak selectivity of Raman spectroscopy allows reliable data interpretation by simple inspection of peak positions. NIR Spectroscopy for Cocrystal Screening. (1) Visual Inspection of Spectra. NIR spectral data from the SAC model system can be seen in (Figure 7), showing the region from 6200 to 4100 cm-1 (combination bands and first overtone region). The fact that spectra are similar for both types of solid material is perhaps the most clear-cut conclusion to be drawn from these results. Assessing the nature of the solid product from the two preparative techniques is a tedious task, as the spectral variability takes place across most of the NIR region. It is not possible to determine rigorously whether the parent and guest molecule exist as a cocrystalline phase or a physical mixture, the latter perhaps containing other solid forms of the API or (hypothetically) the guest molecule. The cocrystal spectrum does seem to contain features that are also evident in the corresponding singlecomponent spectra. Interestingly, there is a band at ∼5925 cm-1, which is not present in the pure reference spectra. To determine whether this absorption band is related to cocrystal formation, it is necessary to be able to interpret the NIR spectra. Band assignment of solid-state NIR spectra, in general, is difficult to

carry out and is associated with high ambiguity. This is due to the anharmonicity of molecular vibrations in the NIR region, which allow overtone absorptions. Thus, a particular vibration can show several absorption characteristics in the NIR spectrum. The absorption band at 5925 cm-1 observed in IND-SAC cocrystal is assigned to first overtone vibrations of either CH or SH bonds. Obviously, this information is not sufficient to confirm the formation of IND-SAC cocrystal. The lack of interpretational ability also applies to the AAC system, in which no cocrystal is formed. Solid outcome generated by liquid-assisted cogrinding is not easily separated visually into spectral contributions from the two single components (Figure 8), It is therefore not possible to provide a clear assessment of the solid-state properties and thus confirm a physical mixture of γ-indomethacin and the guest molecule (as found from XRPD and Raman spectroscopy). For the solution-based method, the risk of drawing erroneous conclusions is lower as the NIR spectrum reveals distinct features of R-indomethacin. Evidently, applying visual inspection (through peak comparison/matching) is of little use when analyzing the NIR data. Other approaches to data mining should therefore be examined. (2) Multivariate Visualization of Spectra. It is common practice to approach the issue of analyzing NIR data by multivariate modeling.24 Aaltonen et al. used PCA to decompose the spectral variation of NIR data acquired from a polymorph screen.18 This made it possible to visually detect spectral changes of the recrystallized product as proof of a new API polymorph or solvate. Unfortunately, use of multivariate analysis on cocrystal data creates further practical problems. These issues are illustrated in Figure 9a and b, showing PCA results from the SAC and AAC model systems, respectively. Besides the solid product form, the two preparative methods and single-component samples, a set of binary mixtures of API and guest molecule have also been included. These additional samples are denoted reference samples and (24) Roggo, Y.; Chalus, P.; Maurer, L.; Lema-Martinez, C.; Edmond, A.; Jent, N. J.Pharm.Biomed.Anal. 2007, 44, 683–700.


7761

Figure 8. SNV-corrected NIR spectra of indomethacin/L-aspartic acid solid product prepared using the two methods along with single component spectra. The Y-axis has been offset to improve the clarity of the spectra.

together constitute a so-called reference space, displaying potential solid-state properties of solid material generated in this particular cocrystal screening (i.e., polymorphic transition of API retained in a physical mixture). All reference samples were manually prepared by geometric mixing, since the anharmonicity of molecular vibrations in the NIR region prevents the calculation of

Figure 9 7762


meaningful addition spectra, as was done for the Raman data. NIR spectral data from the SAC system was decomposed into a total of three principal components (PCs) describing 97% cumulative variance (Figure 9a). Three PCs were required to fully map the dissimilarity between the measured samples. PC1 (63% variance) captures variance related to the ratio between host and guest

Figure 9. (a) Score plot of NIR data for IND-SAC cocrystal as well as 13 reference samples including binary mixtures of indomethacin solid forms and saccharin. The first three principal components explain 97% cumulative variance. The numeric value of a PC3 score for a given sample is indicated by the corresponding color bar on the X-axis. Hence, samples with high content of R-indomethacin have the highest PC3 score values (i.e., between 0.05 and 0.2; pink/purple), while samples with γ-indomethacin have the lowest PC3 values (i.e., ∼-0.1; cyan). Binary mixtures with amorphous content lie between R and γ samples in the PC3 direction (blue). Note: the standard deviation of triplicate measurements falls well within the area of the marker circle. (b) Score plot of NIR data for indomethacin/L-aspartic acid solid product as well as 13 reference samples including binary mixtures of indomethacin solid forms and L-aspartic acid. Note: the standard deviation of triplicate measurements falls well within the area of the marker circle.

molecule, PC2 (23% variance) primarily separates the cocrystalline phase from the remainder of the sample set, while the three API solid forms are distinguished in PC3 (11% variance). Clearly, spectra of IND-SAC cocrystal deviate markedly from the reference space in PC2, thereby disproving formation of any physical mixture. This indirectly suggests the formation of INDSAC cocrystals. NIR data of AAC samples were decomposed into two PCs explaining 95% cumulative variance (Figure 9b). Again, a correct interpretation was only possible when binary mixtures were included in the data set. The sample was produced by liquidassisted cogrinding clusters with the γ-indomethacin and AAC samples (i.e., the starting material), while the sample recrystallized from solution is located in the region spanned by reference samples containing R-indomethacin. This verifies that the spectral change observed for the sample subjected to the solution-based method is in fact related to a solid-state transition of the API and not cocrystal formation. In contrast to Raman spectroscopy, NIR does not carry the disadvantage of subsampling as observed from the clear separation of binary mixtures of varying ratios for both model systems. This is due to the low absorption coefficient of NIR light resulting in a high sampling volume. To summarize, the increased solid-state complexity of the outcome from a cocrystal screening makes implementation of a

NIR/multivariate method challenging. This is in contrast to polymorph screening, in which the sample material only constitutes a single chemical component. It is therefore safe to assume that any spectral changes observed using multivariate modeling are tantamount to polymorphic transformation or solvate formation. The multicomponent nature of a cocrystal makes it impossible to apply a similar approach for positive verification of a cocrystalline phase. CONCLUSION The analytical potential of noninvasive NIR spectroscopy for cocrystal screening was explored using Raman spectroscopy as the comparative method. Model compounds included indomethacin and the guest molecules, saccharin and L-aspartic acid; the former guest molecule is claimed to form a stable cocrystal with indomethacin while the latter is not. The low selectivity of NIR was found to compromise screening efficiency whenever polymorphic APIs were employed, as was the case for the API used here. Therefore, to minimize the risk of false positives, an additional set of reference samples should be added to the data set. These reference samples represent possible solid-state outcomes of the cocrystal screening. Reference mixtures relevant to this study were as follows: two API polymorphs and one amorphous form, each retained in a physical mixture with the Analytical Chemistry, Vol. 80, No. 20, October 15, 2008

7763

respective guest molecule. As shown by principal component analysis, a cocrystal NIR spectrum does indeed differ from that of the corresponding physical mixture(s), and positive verification of cocrystal formation is therefore done on an indirect basis. In contrast, the high selectivity and good interpretational abilities of Raman spectroscopy make it possible to identify spectral contributions from each chemical constituent by a peakwise comparison of single-component spectra (API and guest) and the two-component sample material (API/guest), thus allowing a direct assessment of cocrystal formation including solid-state transitions. For Raman spectroscopy, though, drawbacks such as

7764


a marked difference in Raman activity between API and guest molecule as well as subsampling effects were found to impair the interpretation slightly. The issue of low selectivity of NIR spectroscopy together with the solid-state complexity of sample material produced in cocrystallization experiments should be considered when applying NIR spectroscopy in related screening activities. Received for review June 4, 2008. Accepted August 18, 2008. AC8011329

Near-Infrared Spectroscopy for Cocrystal Screening. A Comparative

Recommend Documents