Confocal Raman Microscope Mapping of a Kofler Melt

DOI: 10.1021/cg1010924 Confocal Raman Microscope Mapping of a Kofler Melt Aalae Alkhalil, Jagadeesh B. Nanubolu, Clive J. Roberts, Jonathan W. Aylott, and Jonathan C. Burley* 2011, Vol. 11 422–430 Received August 20, 2010; Revised Manuscript Received November 23, 2010 ABSTRACT: The spatial distribution of molecular and crystalline entities in a model Kofler melt was characterized using confocal Raman microscopy/spectroscopy. Direct classical least squares and principal component analysis were used to generate hyperspectral maps. The long-standing view that the Kofler method is suitable for rapid and reliable mapping of a binary phase diagram is not supported from our results. Introduction Multicomponent solids are widely discussed in the context of a variety of applications in pharmaceutical, chemical, and material fields.1,2 Co-crystals, in particular, have recently gained a great deal of attention in the pharmaceutical area.3,4 They have been adopted as an approach to modify the properties of single-component materials, which has become a matter of importance in drug development and delivery fields.5-9 Co-crystals can be defined as single-phase multicomponent solid materials, where components coexist in stoichiometric ratios and are linked through noncovalent bonds with the components remaining un-ionized.10 Few techniques have been employed to generate co-crystals, of which solution screening is the dominant method of isolating new crystalline forms. However, this can be complex when applied to co-crystal formation, especially for components of differing solubilities.11,12 This issue can be addressed by conducting the crystallization from a melt, which avoids the use of any solvent. An example of this is the use of the fusion contact protocol by means of hot-stage microscopy.13,14 The concept of fusing mixed compounds based on the contact method was originally described by Lehmann in 1888,15 which was later refined by Kofler16 and called the contact method by McCrone.13 In fact, the use of the fusion contact protocol for screening co-crystals is not particularly widespread and has remained unmodified since McCrone’s work.13 However, it has been recently revisited by a few researchers such as Davis et al.,17 McNamara et al.,18 and Berry et al.19 The basic premise of the Kofler mixed fusion method is to allow two components to meet between a glass microscope slide and a coverslip, and form (by melting) a composition gradient. Subsequently the potential formation of a co-crystal at this zone of melting is investigated by optical microscopy. The fusion method can also be used to generate small seed crystals which can subsequently be used to grow larger crystals from solution (suitable for structure determination by single crystal X-ray diffraction).11 However, the main idea of using hotstage microscopy in Kofler preparation is to screen the melting behavior of two component systems in a single rapid and straightforward experiment, without the necessity to determine their full phase diagram.20 The conventional theory of the *To whom correspondence should be addressed. E-mail: jonathan.burley@ nottingham.ac.uk. pubs.acs.org/crystal Published on Web 01/04/2011 observed phases in an ideal Kofler experiment assumes a linear composition gradient across the zone of mixing after starting with 100% of the respective materials at the peripheral sides (Figure 1a).11,14,19,21,22 The co-crystal formation is potentially achieved from the liquefied mixture at any area within the contact region, provided that the components’ composition exists at a suitable molar ratio for the crystallization of single-phase form (typically 1:1, 1:2, 2:1, or other commensurate ratio). This formation is separated from the components present at the peripheral sides with solidified eutectic mixtures existing at different ratios. Thus, Kofler melt preparation allows a rapid qualitative mapping of the phase diagram, as outlined in Figure 1b. Despite all the benefits of traditional hot-stage microscopy in Kofler mixed fusion preparations, there are some serious limitations of this method in achieving full qualitative and quantitative characterization. One severe limitation is that the hot-stage microphotographs often cannot unambiguously identify the phases present, as their interpretation depends on visual inspection. Therefore, such an investigation is subjective and cannot be used for example to unequivocally identify phases. We propose in this contribution that such imperfections could be overcome by the use of spectroscopic mapping techniques for screening Kofler melts. Because the sample is held between a coverslip and a glass microscope slide, infrared mapping methods such as contact method attenuated total reflectance IR (ATR-IR) are inappropriate as the Kofler melt is not directly accessible. We have therefore employed confocal Raman microscopy, which can be performed in a noncontact, confocal manner,23 in the present study to investigate the spatial distribution of chemical and crystalline entities in a Kofler melt. Raman spectroscopy/mapping is a well-established technique and is widely employed in many fields including pharmaceuticals.24 It is typically used to provide information on the molecular nature of the sample by employing the spectral window 400-4000 cm-1, which covers the energy range of intramolecular vibrations. However, by employing the lower energy phonon-mode bands (10-400 cm-1), it is possible to probe the intermolecular vibrations and thereby obtain detailed information about the solid-state characteristics of the sample, including polymorphism.25,26 Phonon-mode data have recently been employed in generating Raman maps of spatially distributed polymorphs in molecular semiconductors,27 where they proved to be far more sensitive to crystalline packing than the traditional intramolecular bands. In the r 2011 American Chemical Society Article Crystal Growth & Design, Vol. 11, No. 2, 2011 423 Figure 1. Schematics of the traditional component’s position in Kofler melt preparation and its binary phase diagram. Panel (a) is the plot of molecular fraction vs its position in Kofler melt slide, while (b) is the binary phase-diagram of the co-crystal formation. The temperature in the phase diagram is plotted vs the molar fractions of components (A and B, the higher and lower melting components, respectively). It shows the co-crystal formation (C) from stoichiometric (A and B) at 50%. It also displays the melting of (A, B) at 100% of each, Tm (A) and Tm (B), respectively. It indicates the meeting in the liquid phase between component (A) and co-crystal (C), component (B) and the co-crystal (C) at the lowest temperature or eutectic temperature as indicated by asterisks where the intersection occurs in wide and smooth style. Figure 2. Molecular structure of flurbiprofen (a) and nicotinamide (b). present study, we employed confocal Raman spectroscopy/ microscopy to study the molecular and crystallographic nature of a model Kofler melt system through both phononmode and intramolecular bands. Our study provides detailed insight into the applicability of the Kofler method for cocrystal screening and represents the first application of spectroscopic mapping method to a Kofler melt. The phononmode Raman data on the co-crystal are also the first example of the application of this emerging spectroscopic technique to characterize co-crystals. The well-characterized model system flurbiprofen (FBP)nicotinamide (NCT), which was previously investigated by Berry et al. using standard hot-stage microscopy, was selected for our study.19 The characterization in the Berry et al. study was based on standard hot-stage optical microscopy, which provided clear evidence for four separate phases in the FBPNCT system: the co-crystal, FBP, and two NCT polymorphs.19 Thus, we were encouraged to use this system as a model in our spectroscopic mapping study. FBP is used as an active pharmaceutical ingredient (API) and NCT is the co-crystallizing agent. FBP (Figure 2a) falls in the carboxylated group of nonsteroidal anti-inflammatory drugs, while NCT (Figure 2b) is a vitamin B3 derivative. NCT is generally regarded as safe and suitable for human use.28 It possesses a pyridyl ring and amide group in its structure. Both groups allow NCT to serve as a complementary cofactor and a solubility enhancing agent. Experimental Section Raman spectral data were acquired using a LabRAM HR system (Horiba Jobin Yvon), operating with an 800 mm path-length, a 600 lines/mm grating, a 532 nm laser source, and a 50 objective lens and a Synapse thermo-electrically cooled CCD. Temperature control was achieved using a Linkam LTS350 hot-stage. The system was calibrated before acquiring measurements on a standard silicon, as well as the Rayleigh line. The silicon Raman peak was within (1 pixel of 520.7 cm-1. The grating was fixed to allow relatively rapid mapping, at the expense of a slightly restricted spectral range. Such combination improves the quality of data achieved through this instrument and allows for an access to the phonon-mode spectral window (30-400 cm-1) as well as the higher wavenumber region (4004000 cm-1) which provides information on the traditional intramolecular bands. Thereby, both the intramolecular and intermolecular spectral information could be obtained. The mapping was automated by an XY stage and set up to cover a grid of (1550 μm  1625 μm) at 25 μm steps. 4030 Raman spectra were acquired based on an acquisition time of 3 s per spectrum. FBP and NCT were purchased from Sigma-Aldrich (Poole, UK) and were used as received. X-ray powder diffraction (XRPD) confirmed that they were of form I crystalline state when compared to patterns obtained from Cambridge Structural Database (library Reference codes are “FLUBIP”, and “NICOAM03”, respectively).29,30 They were also found to melt as per literature by differential scanning calorimetry (DSC) (114.01 and 128.24 °C, respectively).31,32 Reference samples of the two FBP polymorphs were prepared by solution methods and characterized by X-ray powder and single crystal diffraction, and Raman spectroscopy. The single crystal structure of FBP form II was achieved in an attempt to prepare FBP-NCT cocrystal from solution (Figure S1, Supporting Information). Solvent was analytical grade supplied by Fisher Scientific (Loughborough, Leicestershire, UK). The process of Kofler melt preparation was carried out by placing a small amount of NCT on a glass microscopic slide under a coverslip. This was heated until melting occurred at around 130 °C, and then allowed to solidify. FBP was later placed at the opposite edge of the coverslip and melted at around 115 °C. Once melted, it was drawn by capillary action under the coverslip dissolving the juxtaposed part of the NCT. The whole preparation was then allowed to cool and kept at RT to solidify for ease of sample handling. All of our experiments were performed at room temperature; therefore, the entire samples and the various interfaces identified are static. As a premapping procedure, Raman spectra were taken for the melted starting materials (FBP form I and NCT) and from their equimolar melted mixture. These Raman spectra were taken under the same conditions as for the Kofler preparation and therefore used as references. Raman spectra of the starting materials NCT and FBP form I, FBP form II, and co-crystal at the zone of mixing were acquired (Figure S2, Supporting Information) and used for assigning principal components’ loadings to their original components. Spectral data sets were subjected to baseline correction and mean normalization before carrying out direct classical least squares (DCLS) analysis of the melted region. DCLS maps were generated by employing a limited number (n) of selected spectra to form a linear combination, which was then used to model all of the spectra in the Raman spectral image. The modeling is achieved through a best-fit methodology (eq 1). The sum of the (n) components would not often be able to model 100% of a given spectral pixel in the image which is expressed by an error (ε) as shown in the eq 1. Both the molecular composition and crystalline form are of interest in a Kofler melt and 424 Alkhalil et al. Crystal Growth & Design, Vol. 11, No. 2, 2011 Figure 3. Fusion method preparation of flurbiprofen-nicotinamide and single Raman spectra. Shown in (a) is Kofler preparation after cooling which includes the peripheral regions, flurbiprofen (left) and nicotinamide (right), and the zone of mixing at the center in between; all are indicated by arrows, while (b) is Raman spectra of NCT (green), co-crystal (blue), and FBP (red). were investigated through employing the selected numbers (n) in particular spectral windows for the analysis. Setting n = 2 would correspond to FBP and NCT molecules, whereas setting n=3 refers to the crystalline forms FBP, NCT, and a potential co-crystal. In addition to selecting a given number of spectra with which to create the DCLS model, there is also a choice of spectral window selection in which DCLS modeling was performed; in particular, the phonon-mode region (30-400 cm-1), the molecular stretching region (400-1721.6 cm-1), and the entire wavenumber region (30-1721.6 cm-1). Y ðvmÞ ¼ K  X ðcmÞ þ ε ð1Þ Equation 1 is a classical least-squares fitting model. Through this equation, a number of spectra (m) at a specific v-wavenumber of composition (c), arranged into Y matrix (v  m), are classified according to each representative spectrum (n) into a column of concentration matrix X(c  m) with a matrix of sensitivity K and an error ε. Maps were also generated using principal component analysis (PCA) in the spectral regions mentioned previously in Matlab 7.6.0 (R2008a). PCA simplifies the data into sets of variables, which are linearly correlated with the original data. It characterizes the variances in data by creating uncorrelated axes leading to transformation in the original coordinates. The axis that gives the highest variance between the original data refers to the first score, the one with consequent variance is the following one, and so on.33 Raman spectral profile data were split into individual spectra, baseline corrected under LabSpec5 software, and imported into Matlab 7.6.0 (R2008a) in text format. Each variable was divided by the standard deviation of its Raman spectrum, and all original data were mean normalized prior to PCA analysis. The results were rescaled to represent their intensities for use in red-green-blue (RGB) color-map as pixels with higher intensity can mask less intensity changes elsewhere. Images areas with positive loadings were assigned red derivative color while negative ones were in blue. The first three principal components (PCs) were selected in this study having statistically significant values accounting for >93% of variances (Figure S3, Supporting Information). Loadings were compared with Raman spectra of NCT, FBP form I and II, and the spectrum extracted from the zone of mixing. Results and Discussion Kofler Preparation. Author: Please verify that the changes made to improve the English still retain your original meaning.Several attempts were carried out in order to achieve the required preparation, represented by holding the sample for a time above the eutectic temperature before being fully solidified. This difficulty in crystallising the co-crystal in the Kofler preparation is thought to be due in large part to the stochastic nature of the nucleation process. The formation of a suitable Kofler melt for analysis was confirmed optically (Figure 3a). Inspection shows clearly the existence of at least three crystalline phases in the Kofler melt, assigned as NCT, FBP, and co-crystal. This was also chemically proven when investigating their Raman spectra acquired from the three optically different regions (Figure 3b). Each of these spectra, referring to NCT, FBP, and co-crystal from Kofler preparation, was in good agreement with the referenced samples prepared separately from a fusion procedure in which the composition gradient was not generated (Figure S4, Supporting Information). In contrast to Berry et al.,19 we did not observe any polymorphism in NCT in the small area sampled in our experiment. The generation and the screening of polymorphs generally require a statistically significant number of experiments, so the nonappearance of second polymorph of NCT in our results is not an issue for concern in this report. Raman Mapping. Raman mapping of the NCT-FBP Kofler melt preparation was conducted following the optical and spectral confirmation of the preparation and its data were analyzed by means of DCLS and PCA. For the DCLS analysis, a number (n) of representative spectra exhibiting similar features to the pure components were needed as an input for generation of the DCLS map. Such a requirement was achieved by extracting spectra from the mapping data and comparing them to the reference samples of NCT, FBP, and co-crystal prepared separately from the melt. Figure S4, Supporting Information shows the possibility of using any of the spectra extracted from the three optically different regions as representative models after confirming their similarity to the referenced samples. Two (n) selections, n = 2 and 3, were designated to probe the molecular and crystalline species, respectively. The selection n = 2 refers to the starting (molecular) materials NCT and FBP, while n = 3 alludes to the expected crystalline materials (the emerging phase (co-crystal) in addition to both crystalline starting materials). In addition to selecting the number of DCLS components, three selected spectral windows were employed in these analyses, in order to address the spatial distribution of molecular and crystalline entities. The traditional intramolecular spectral window (400-1721.6 cm-1) is expected to be most sensitive to the molecular composition of the samples, whereas the phonon-mode data (30-400 cm-1) are expected to be more sensitive to the crystallographic nature of the samples. We also present an analysis of the entire spectral range for comparison. Maps generated from a DCLS analysis with n = 2 (i.e., maps achieved from reference spectra for NCT and FBP only) in the three spectral regions are presented in Figure 4. Article Crystal Growth & Design, Vol. 11, No. 2, 2011 425 Figure 4. DCLS maps at two number of selection n = 2. Two-component scores are given in a-c, and their models are in d-f. Maps generated in the phonon-mode region (30-400 cm-1) are given in a and d, the molecular stretching region (400-1721.6 cm-1) in b and e, and the entire spectral range in c and f. The colors in the peripheral regions refer to the compositions of the higher and lower melting components (green and red, respectively) as shown from their models. All three maps exhibit at least two well-defined regions which are separated by vertical boundaries in Figure 4a-c. The regions themselves are indicated by colors in the Figure 4a-c and their representative spectra are in Figure 4d-f. The comparison of the optical image (Figure 3a) with the DCLS maps at n=2 indicates that the left-hand region in the DCLS maps corresponds to FBP plus co-crystal, and the right-hand to NCT only. DCLS maps conducted at the three spectral regions are in general rather similar. However, some differences are apparent on closer inspection. For example, there is an evidence of a third region running vertically down the middle of the image in the map generated from the phononmode spectral window data (Figure 4a), and to a lesser extent in the map generated from the entire spectral window (Figure 4c). This region corresponds to the co-crystal, which was again deducted from the comparison to the optical image in Figure 3a. Note that a reference spectrum for the co-crystal was not included in the input for map generation, and thus it is slightly surprising that the co-crystal can be identified from these maps. The co-crystal in these maps is manifested by a high loading on the error term ε in eq 1 (Experimental Section), rather than a direct loading on either of the two input spectra. In contrast to the phonon-mode map, the co-crystal is not readily apparent from inspection of the map generated from the intramolecular spectral range (Figure 4b). Overall, for the n=2 DCLS map generation, the presence of the co-crystal is most clear in the phonon-mode data, and this is in accord with the phonon-mode data being more sensitive to the nature of the crystalline phase(s) than other spectral ranges. Figure 5 presents the results of generating DCLS maps with n=3 (i.e., the input of spectra herein refers to FBP, cocrystal, and NCT). The three spectral ranges were similarly employed to generate separate maps. All three maps exhibit three distinct regions which correspond (from left to right) to FBP, co-crystal, and NCT. Again the maps are similar in broad terms, but differences emerge on detailed inspection. In all three maps, the interface co-crystal/NCT is very distinct, with the FBP/co-crystal interface being less discrete. This clearly indicates that the composition gradient is different in nature at these two interfaces. The FBP/co-crystal interface is less distinct in the map generated from the molecular data than in the phonon-mode data (the molecular conformation in the FBP-NCT cocrystal appears to be similar to those in both FBP and NCT (Figure 5e) unlike the differences in the intermolecular interactions as manifested in the phonon region (Figure 5d). The obvious interpretation of this is that the molecular composition gradient is relatively smooth, whereas the crystalline composition gradient is relatively sharp. This is expected for a Kofler melt, in which a molecular composition gradient is used to isolate a co-crystal of a particular composition. The co-crystal is expected only to form in a limited composition range, and the results observed are fully in accord with this. The differences in the nature of the interfaces in the FBP/ co-crystal and, the co-crystal/NCT was commonly noticed in analyses generated for all spectral windows. This suggested the possibility of investigating the composition gradient across the Kofler melt preparation, an issue that has not been addressed in previous work. Figure 6a-c shows the averaged percentages of the existing components, taken from eight horizontal lines across the preparation, calculated from DCLS when n = 3 at the phonon region, the molecular region, and the whole wavenumber window, respectively. The relative amounts of the components do not correspond directly with the molar quantities due to calibration issues including differing Raman scattering cross sections of the 426 Crystal Growth & Design, Vol. 11, No. 2, 2011 Alkhalil et al. Figure 5. DCLS maps at three number of selection n = 3. Three-component scores are given in a-c, and their models are in d-f. Maps generated in the phonon-mode region (30-400 cm-1) are given in a and d, the molecular stretching region (400-1721.6 cm-1) in b and e, and the entire spectral range in c and f. The colors in the peripheral regions refer to the compositions of the higher and lower melting components (green and red, respectively), while the one at the zone of mixing (blue) refers to the co-crystal as shown also in their models. Figure 6. Composition gradient deduced from DCLS using three components selection n = 3 in three spectral windows. Panels (a), (b), and (c) show the analysis generated from the phonon-mode data, the molecular data, and from the entire spectral data, respectively. various components, which is beyond the scope of the present work to address. However, the relative amounts of material from our analysis do provide a good guide to the nature of the composition gradient if not the exact values of the composition. Analysis performed at the three spectral regions (Figure 6a-c) all show that the composition gradient of both starting components was not ideally linear, especially toward the higher-melting component (NCT). This is clearly noticed from the obvious sharp descent in NCT concentration and the sudden rise of co-crystal concentration. However, there is a gradual decrease/increase in components’ concentration at the FBP/co-crystal interface. The reason for this can be deduced by considering the manner in which the Kofler melt is prepared and studied. Conceptual perspective on such findings is proposed by taking into consideration the sequential events of Kofler preparation (Figure 7 I-IV): 1 The highest melting material (NCT) was offset to one side, once it solidified after being melted (Figure 7 I); 2 Later, the lower melting material (FBP) was melted when brought into contact with the juxtaposed part of the higher melting material (NCT) (Figure 7 II); 3 The lower melting material dissolved some solidified parts of the higher melting component NCT at the contact area, while the rest of the parts remained inaccessible (Figure 7 III). This is analogous to the effect of tides on the seashore, whereas the powerful sea waves erode only the facing coastal features, depositing the washed sands offshore. This phenomenon was clearly demonstrated from DCLS results, which revealed the Article Crystal Growth & Design, Vol. 11, No. 2, 2011 427 Figure 7. Schematic of proposed sequence of events in the formation of a Kofler melt in (I-IV). It shows the consequential steps for preparing Kofler sample, whereas the higher melting component is (A) shown in green, the lower melting component is (B) shown in red, and the co-crystal is (C) shown in blue. Figure 8. Schematics of our perspective about ingredient’s position in Kofler melt preparation and its binary phase diagram, showing the restricted access to the higher melting component region. Panel (a) is the plot of molecular fraction vs its position in Kofler melt slide, while (b) is the binary phase-diagram of the co-crystal formation. The temperature in the phase diagram is plotted vs the molar fractions of components (A and B, the higher and lower melting components, respectively, while C is the co-crystal). The intersection occurs in a sharp form only in the melting component section as shown by the green line. discontinuity of the component distribution in NCT region (Figures 4a-c and 5a-c). Thereby, the higher melting component forms an area with a very distinct border at the interface with the zone of mixing. 4 Finally, the higher melting material NCT portions already dissolved to the saturation point in the liquefied lower melting component FBP would crystallize forming a stripe of co-crystals along the contact area (Figure 7 IV). Some detached NCT particles and unsolidified co-crystal might diffuse in the lower melting material (FBP) (Figure 7 IV), depending on how fast the lower melting component crystallizes. This is inherently difficult to control, and it is correlated to the solubility and the kinetics of dissolution process. The obviously different nature of the two interfaces, observed for the first time in this work, refines the traditional vision of a linear composition gradient in Kofler preparation. Our results (Figures 6a-c) unambiguously demonstrate that a linear composition gradient does not form, and indeed on detailed consideration of the process of forming a Kofler melt, it is not expected to form. This is illustrated schematically in Figure 8a, in which the traditional vision of the Kofler melt preparation is compared with the actual situation we find in this work. It shows the direct interface of the higher melting point solid with the rest of the preparation, instead of a linear composition gradient causing to a sharp change in the composition. This finding has important implications for the use of the Kofler melt as a screening method for co-crystal formation. In particular, the step change in the composition of the preparation by the highest melting component will restrict access to this part of the phase diagram (Figure 8b, intermittent line). Thus, any potential co-crystals with compositions rich in the highest-melting component may not be experimentally accessible. The composition range which is rendered inaccessible will depend on a number of factors, the most important of which is likely to be the amount of time spent by the lower-melting liquid in contact with the highermelting solid. The longer this time is, the more the chance is for the higher melting solid to dissolve into the lower melting liquid, and the more likely for the sharp change in composition at the interface to be reduced. Thus, our DCLS findings represent an important step forward in understanding the crystal growth in Kofler melts, which would be difficult to quantify without the use of the Raman microscopy technique. Finally, the PCA was the second chemometric method employed for Raman mapping analysis. Such an approach derives a number of independent linear combined variables that has sufficient information about the original chemical data without the interference of operators, unlike DCLS which requires referenced spectra for linearly modeling data. The first three principal components were selected for this study, whose scores were heavily loaded by the presence of compounds in the three optically different regions, as shown in Figure 9aI-cI. These components displayed a homogeneous distribution of pure composition in the higher melting region, which forms a very distinct boundary once reaching the zone of mixing. In contrast, PCA results highlighted the nonuniformity in the distribution of components in the 428 Crystal Growth & Design, Vol. 11, No. 2, 2011 Alkhalil et al. Figure 9. PCA analysis results. The first three PCs scores are shown in aI, bI, and cI, their corresponding loadings in aII, bII, and cII. Shown are (d) the original Raman spectra of NCT (black), co-crystal (green), FBP I (red), and FBP II (blue) (from top to the bottom). RGB color-map image from PC1, 2, and 3 is shown in (e). lowest melting region and its unclear boundary with the zone of mixing region. The homogeneous composition in the higher melting region was clearly confirmed by studying the first principal component (PC1). Figure 9aII shows that PC1 was positively weighted only by NCT, which was verified by the comparison of the NCT Raman spectrum (Figure 9d) and the PC1 loadings (Figure 9aII). Alternatively, the lower melting region displayed heterogeneity in components’ composition. Such an issue was proven by investigating PCs’ loadings. Figure 9bII,cII shows that components’ weight on PC2 and PC3 refer to both FBP and co-crystal. This was confirmed when comparing their positive and negative loadings to the referenced Raman spectra of co-crystal and FBP, respectively. However, the loadings exhibited some positive weights, specifically on PC2, which does not initially match with any of the referenced spectra. These loadings occur strongly in the left-hand side of the images, which does not correspond to FBP form I and alerted us to the likelihood of polymorphism in this component. In this context, polymorphism in FBP has been reported in the literature. Three different polymorphs of FBP are known. Form I is the most stable form, form II could be induced by heteronucleation approaches with polymers, and it could also be prepared from seeded solutions. However, it is less stable than form I as it transforms to form I at around 90 °C. The third form is the least stable, although it is stable between a slide and a coverslip. Moreover, it has been reported that the three forms can be simultaneously crystallized from melt-quenched FBP form I.31,34,35 Figure S5, Supporting Information confirms the similarity of reference spectra for FBP forms I and II to those extracted from two different pixels in the Kofler melt preparation. Article Crystal Growth & Design, Vol. 11, No. 2, 2011 It clearly indicates that polymorphism of FBP does indeed occur in our Kofler preparations. Thus, PCA analysis was an effective exploratory tool for data interpretation especially when compared to DCLS. It revealed unexpectedly the presence of polymorphism in FBP, which has not been noticed in our DCLS approaches. The apparent lack of DCLS selectivity to the polymorphism of FBP is in fact purely an operator issue - we had not included input spectra for the two polymorphs of FBP into the DCLS map generation, as we did not expect to observe it. With the benefit of hindsight, even the optical image (Figure 3a) presents some evidence of FBP polymorphism through the lower visual homogeneity of the FBP than the NCT, although this lower visual homogeneity is easy to mistake for larger differences in crystal habit and orientation when compared to NCT. Although NCT is well-known to exhibit polymorphism (and this was observed by Berry et al.19), we observed no evidence for the presence of more than one crystalline form of this molecule in our experiment. This is not surprising as our experiment only samples an extremely small subset of experimental conditions (temperature, pressure, sample volume, etc.) and it is not expected to act as a tool for screening polymorphism. Nucleation of polymorphs is in any case a stochastic process and therefore not always reproducible between apparently identical experiments.36 Finally, a RGB color map image (Figure 9e) was created by overlapping the first three scores (being assigned by the anticipated/unexpected compounds and having variances of >93%) (Figure S3, Supporting Information). This demonstrates the localization of the components in three regions, which was also noticed from DCLS scores at n= 3. Overall, both multivariate approaches used in our study were invaluable. DCLS analysis was beneficial for yielding quantitative information. However, this needs to be preceded by a knowledge of all existent forms, which is not required for PCA analysis. Thus, both chemometric techniques yielded advantageous interpretation of the hyperspectral data as they could be served to quantify and characterize all crystal forms in the model Kofler melt. Conclusion The work presented in this article outlines the first use of confocal Raman microscopy to characterize Kofler melt preparations. This was shown to be valuable in generating detailed information on the investigated phases and provided qualitative information about the newly formed phases (co-crystal and component’s polymorphs). Detailed quantitative and qualitative information from Raman mapping were achieved with the use of chemometric approaches. In this contribution, two chemometric methods (DCLS and PCA) were explored to investigate the spatial distribution of the components, and restricted spectral ranges were employed to characterize the molecular and crystalline natures of the Kofler preparation. The use of DCLS analysis provided a new perspective regarding the composition gradient present in Kofler melt preparations. It was shown that the composition gradient is not linear as a function of position as had been assumed until now. Rather, a sharp interface exists at the interface of the highest melting component and the cocrystal. This sharp interface restricts access to the binary phase diagram in Kofler preparations at compositions rich in the highest melting component. This may affect the utility of the Kofler melt as a general-purpose method for screening 429 co-crystal formation. It is now advisible, according to our observation, to leave Kofler preparation as long as possible at high temperatures ; over the melting point of both components ; in order to allow the highest melting solid to dissolve in the liquid of the lower melting component, and thus to reduce as far as possible this inaccessible region of the binary phase diagram. In contrast to DCLS, PCA has not only confirmed the distribution of components but also revealed the presence of both FBP polymorphic forms (forms I and II), which had not been deduced from the traditional method of optical observation (either by ourselves, or in the previous work of Berry et al.) nor from DCLS modeling. With the benefit of hindsight provided by the PCA results, the polymorphism of FBP could be noted in the optical images. It is also worth mentioning the sensitivity of screening with respect to the crystalline components which was strongly enhanced by considering the phonon region or phonon-mode data (30-400 cm-1). Raman screening at this region is a highly promising method for investigating in situ crystallization, polymorphs, co-crystals, and solid dosage forms. This is attributed to its significant sensitivity in probing the crystalline structure of molecular compounds. In conclusion, the use of Raman mapping was advantageous for investigating the nature and composition of phases in Kofler preparation which was nearly impossible in the past. From this work, we know that coupling the hot-stage with Raman microscopy provides new insight into the use of Kofler method to screen co-crystal formation and investigate the composition of ingredients. Acknowledgment. We thank the Nottingham Nanotechnology and Nanoscience Centre (NNNC) for providing access to the Raman microscope and the East Midland Development Agency (EMDA) for funding this equipment. A.A. thanks Faculty of Pharmacy, Damascus University for funding her Ph.D. at the University of Nottingham. J.B. and J.N. thank the EPSRC for support under Grant EP/G038740/1. Supporting Information Available: Single X-ray crystal of FBP form II obtained from solution in an attempt to prepare co-crystal using acetonitrile. It was confirmed to be form II when compared with the calculated X-ray pattern taken from the Cambridge Structural Database (CSD) (Figure S1). This material is available free of charge via the Internet at http://pubs.acs.org. References (1) Monissette, S. L.; Almarsson, O.; Peterson, M. L.; Remenar, J. F.; Read, M. J.; Lemmo, A. V.; Ellis, S.; Cima, M. J.; Gardner, C. R. Adv. Drug Delivery Rev. 2004, 56, 275–300. (2) Vishweshwar, P.; McMahon, J. A.; Bis, J. A.; Zaworotko, M. J. J. Pharm. Sci. 2006, 95, 499–516. (3) Shan, N.; Zaworotko, M. J. Drug Discovery Today 2008, 13, 440–446. (4) Almarsson, O.; Zaworotko, M. J. Chem. Commun. 2004, 1889– 1896. (5) Schultheiss, N.; Newman, A. Cryst. Growth Des. 2009, 9, 2950– 2967. (6) Childs, S. L.; Chyall, L. J.; Dunlap, J. T.; Smolenskaya, V. N.; Stahly, B. C.; Stahly, G. P. J. Am. Chem. Soc. 2004, 126, 13335– 13342. (7) Basavoju, S.; Bostrom, D.; Velaga, S. P. Pharm. Res. 2008, 25, 530–541. (8) Trask, A. V.; Motherwell, W. D. S.; Jones, W. Cryst. Growth Des. 2005, 5, 1013–1021. (9) Hickey, M. B.; Peterson, M. L.; Scoppettuolo, L. A.; Morrisette, S. L.; Vetter, A.; Guzman, H.; Remenar, J. F.; Zhang, Z.; Tawa, M. D.; Haley, S.; Zaworotko, M. J.; Almarsson, O. Eur. J. Pharm. Biopharm. 2007, 67, 112–119. 430 Alkhalil et al. Crystal Growth & Design, Vol. 11, No. 2, 2011 (10) Aakeroy, C. B. Acta Crystallogr. Sect. B-Struct. Sci. 1997, 53, 569–586. (11) Blagden, N.; Berry, D. J.; Parkin, A.; Javed, H.; Ibrahim, A.; Gavan, P. T.; De Matos, L. L.; Seaton, C. C. New J. Chem. 2008, 32, 1659–1672. (12) Chiarella, R. A.; Davey, R. J.; Peterson, M. L. Cryst. Growth Des. 2007, 7, 1223–1226. (13) McCrone, W. C. Fusion Methods in Chemical Microscopy; Interscience Publishers: New York, 1957. (14) McCrone, W. C. Microchim. Acta 1951, 38, 476–486. (15) Lehmann, O. Molecularphysik; Engelmann: Leipzig, 1888. (16) Kofler, L.; Kofler, A. Thermal Micromethods for the Study of Organic Compounds and Their Mixtures; Wagner: Innsbrook, 1952; translated by McCrone, W. C.; McCrone Research Institute: Chicago, 1980. (17) Davis, R. E.; Lorimer, K. A.; Wilkowski, M. A.; Rivers, J. H. In Crystals in Supramolecular Chemistry; Beatty, A. M., Ed.; American Crystallographic Association, Inc.: Chicago, IL, 2004; pp 14-23. (18) McNamara, D. P.; Childs, S. L.; Giordano, J.; Iarriccio, A.; Cassidy, J.; Shet, M. S.; Mannion, R.; O’Donnell, E.; Park, A. Pharm. Res. 2006, 23, 1888–1897. (19) Berry, D. J.; Seaton, C. C.; Clegg, W.; Harrington, R. W.; Coles, S. J.; Horton, P. N.; Hursthouse, M. B.; Storey, R.; Jones, W.; Friscic, T.; Blagden, N. Cryst. Growth Des. 2008, 8, 1697–1712. (20) McCrone, 1951:14 “the entire composition diagram is shown in the composition gradient”; Berry et al., 2008:19 “This zone is then comparable to the composition of the binary phase diagram for two components.” (21) Berry et al., 2008:19 “with one side of the mixing zone representing 100 % of one component and the other side 100 % of the other component, and with a concentration gradient across the zone.”; Davis et al., 2004:17 “At some point in the mixing zone, there should (22) (23) (24) (25) (26) (27) (28) (29) (30) (31) (32) (33) (34) (35) (36) be a 1:1 ratio of the two components. Moving from this point toward either pure component, the proportion of that component increases. Thus, a composition gradient exists across the mixing zone.”; and pictorially Kuhnert-Brandstatter, Figures 6 and 7, Ref 22. Kuhnert-Brandst€atter, M. Pure Appl. Chem. 1965, 10, 133–144. McCreery, R. L. Raman Spectroscopy for Chemical Analysis; John Wiley & Sons, Inc.: New York, 2000. Sasic, S.; Yukihiro, O. In Pharmaceutical Applications of Raman Spectroscopy; Sasic, S., Ed.; John Wiley & Sons, Inc., New York, 2008. Day, G. M.; Zeitler, J. A.; Jones, W.; Rades, T.; Taday, P. F. J. Phys. Chem. B 2006, 110, 447–456. Al-Dulaimi, S.; Aina, A.; Burley, J. CrystEngComm 2010, 12, 1038– 1040. Brillante, A.; Bilotti, I.; Della Valle, R. G.; Venuti, E.; Girlando, A. CrystEngComm 2008, 10, 937–946. U.S. Food and Drug Administration, in Database of Select Committee on GRAS Substances Reviews, (SCOGS) http://www. accessdata.fda.gov/scripts/fcn/fcnDetailNavigation.cfm? rpt=scogsListing&id=221, 2006. Batsanov, A. S.; Howard, J. A. K., private communication, 2005 (cited in Cambridge Structural Database). Miwa, Y.; Mizuno, T.; Tsuchida, K.; Taga, T.; Iwata, Y. Acta Crystallogr. Sect. B: Struct. Sci. 1999, 55. Henck, J. O.; Kuhnert-Brandstatter, M. J. Pharm. Sci. 1999, 88, 103–108. Hino, T.; Ford, J. L.; Powell, M. W. Thermochim. Acta 2001, 374, 85–92. Clark, D.; Sasic, S. Cytometry Part A 2006, 69A, 815–824. Grzesiak, A. L.; Matzger, A. J. J. Pharm. Sci. 2007, 96, 2978–2986. Lacoulonche, F.; Chauvet, A.; Masse, J. Int. J. Pharm. 1997, 153, 167–179. Dunitz, J. D.; Bernstein, J. Acc. Chem. Res. 1995, 28, 193–200.