Probing Pharmaceutical Mixtures during Milling - ACS Publications

Article Cite This: Mol. Pharmaceutics XXXX, XXX, XXX-XXX

pubs.acs.org/molecularpharmaceutics

Probing Pharmaceutical Mixtures during Milling: The Potency of Low-Frequency Raman Spectroscopy in Identifying Disorder Greg Walker,† Philipp Römann,† Bettina Poller,† Korbinian Löbmann,# Holger Grohganz,# Jeremy S. Rooney,‡ Gregory S. Huff,‡ Geoffrey P. S. Smith,‡ Thomas Rades,# Keith C. Gordon,*,‡ Clare J. Strachan,*,§ and Sara J. Fraser-Miller‡ †

School of Pharmacy, University of Otago, Dunedin, New Zealand Department of Pharmacy, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark ‡ Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Chemistry, University of Otago, Dunedin, New Zealand § Division of Pharmaceutical Chemistry and Technology, Faculty of Pharmacy, University of Helsinki, Helsinki, Finland #

S Supporting Information *

ABSTRACT: This study uses a multimodal analytical approach to evaluate the rates of (co)amorphization of milled drug and excipient and the effectiveness of different analytical methods in detecting these changes. Indomethacin and tryptophan were the model substances, and the analytical methods included low-frequency Raman spectroscopy (785 nm excitation and capable of measuring both low- (10 to 250 cm−1) and midfrequency (450 to 1800 cm−1) regimes, and a 830 nm system (5 to 250 cm−1)), conventional (200−3000 cm−1) Raman spectroscopy, Fourier transform infrared spectroscopy (FTIR), and X-ray powder diffraction (XRPD). The kinetics of amorphization were found to be faster for the mixture, and indeed, for indomethacin, only partial amorphization occurred (after 360 min of milling). Each technique was capable of identifying the transformations, but some, such as low-frequency Raman spectroscopy and XRPD, provided less ambiguous signatures than the midvibrational frequency techniques (conventional Raman and FTIR). The low-frequency Raman spectra showed intense phonon mode bands for the crystalline and cocrystalline samples that could be used as a sensitive probe of order. Multivariate analysis has been used to further interpret the spectral changes. Overall, this study demonstrates the potential of low-frequency Raman spectroscopy, which has several practical advantages over XRPD, for probing (dis-)order during pharmaceutical processing, showcasing its potential for future development, and implementation as an in-line process monitoring method. KEYWORDS: amorphous, coamorphous, milling, Raman, low-frequency Raman, infrared, indomethacin, tryptophan

■

the term “coamorphous” is now commonly used to describe these systems. Suitable small molecule excipients include citric acid and various amino acids.3 In some cases, two drug molecules have also been combined.6,7 The stabilizing mechanisms are, in principle, the same as those of the polymeric excipients. However, the coamorphous systems may show some advantages over polymer-based dispersions, including reduced hygroscopicity and higher drug loadings. In addition to excipient selection, the method of manufacturing (co-)amorphous systems is important. The method (and its processing parameters) can affect both structure and mobility of the amorphous material,8 which in turn affects the pharmaceutically relevant critical quality attributes of physical stability, solubility, and dissolution rate. Preparation methods

INTRODUCTION The amorphous form is an increasingly popular formulation approach for increasing the apparent solubility and dissolution of poorly water-soluble drugs.1,2 Although the amorphous form of a given material is thermodynamically unstable and susceptible to crystallization, sufficient kinetic stability may, in practice, be obtained (for the period the formulation is expected to be stored and used) through optimized selection of both stabilizing excipients and preparation method. With respect to stabilizing excipients, hydrophilic polymers, within which the drug is molecularly dispersed to form an amorphous glass solution, is the most established approach with several such products on the market. In these systems, crystallization is inhibited through (i) the antiplasticizing effect of the polymers, (ii) specific bonding interactions (e.g., Hbonding) between drug and polymer, and (iii) the mechanical hindrance to crystal growth associated with molecular level mixing.3−5 More recently, small molecule excipients have also been found capable of stabilizing amorphous drug systems, and © XXXX American Chemical Society

Received: Revised: Accepted: Published: A

September 14, 2017 October 30, 2017 November 1, 2017 November 1, 2017 DOI: 10.1021/acs.molpharmaceut.7b00803 Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Article

Molecular Pharmaceutics

spectroscopy and its comparison to other conventional spectroscopic methods (IR and Raman), as well as XRPD.

best suited to industrial scale production include spray-drying, melt extrusion, and milling. Milling avoids the use of solvents and high temperatures. However, challenges include variation in the time to render the system sufficiently disordered, as well as differences between analytical methods (spectroscopic, thermal, and X-ray based) with respect to the level of detected disorder at different milling times. For example, recently it has been shown that while drugs become progressively more disordered over time until a maximum level of disorder is reached, the material may appear to become disordered more rapidly with X-ray powder diffraction (XRPD) than Raman spectroscopy due to the scale at which they detect disorder with XRPD being a lattice level technique, while Raman spectroscopy probes smaller-scale inter- and intramolecular bonding.9 Raman spectroscopy is sensitive to solid state structure since the associated differences in molecular arrangement and/or conformation with different solid state forms are reflected as changes in intra- and intermolecular vibrational behavior. Raman spectroscopy exhibits some advantages over X-ray powder diffraction (XRPD) for solid state analysis in pharmaceutics, including more flexible noncontact sampling setups, lower instrumental costs, easier implementation in processing environments, and a smaller preferred orientation effect. While orientation effects are observed in crystals using mid- and low-frequency Raman spectroscopy,10 these are minimized in a randomly orientated powder mixture. Raman spectroscopy is now widely exploited to identify and quantify solid state forms of drugs, during, for example, polymorph screening,11,12 processing operations, either in unit-based13 or continuous14 manufacturing setups, storage,15 and dissolution.16,17 In the pharmaceutical setting, Raman spectroscopy typically involves detection in the mid- to high-Raman shift regions. Depending on the detection setup used, this varies somewhere within the 150 to 4000 cm−1 region. In organic materials, this region corresponds to various vibrational modes that are principally intramolecular. Changes in the intermolecular environment are often reflected as reasonably subtle changes in Raman shifts and intensities. As a result, Raman spectra of different (crystalline) solid state forms of molecular compounds are usually similar but not identical. The Raman spectra of amorphous forms also exhibit some spectral differences, as well as peak broadening due to a distribution of molecular conformations and interactions.18 Quantification of these differences requires the employment of multivariate analysis methods such as partial least-squares regression. The low-frequency region of Raman spectroscopy (∼10−250 cm−1) is interesting for solid state analysis of pharmaceutical materials because it probes vibrational modes that are principally intermolecular, making it inherently more sensitive to solid state structure than its mid- and high-frequency counterparts. A number of studies have highlighted the effectiveness of low-frequency Raman in this respect.19 In addition, studies have looked at identifying different polymorphic forms of drugs such as indomethacin,20 caffeine,21 and carbamazepine.22 Low-frequency Raman spectroscopy has also been utilized for monitoring solid state changes such as phase changes during tablet manufacturing23 and monitoring the crystallization of active pharmaceutical ingredients (APIs) using bivariate20 and multivariate24 methods. This study reports the first analysis of a coamorphous mixture formation during milling using low-frequency Raman

■

MATERIALS AND METHODS The amino acid L-tryptophan (TRP) was purchased from Sigma-Aldrich (St. Louis, MO, USA). The API, indomethacin (IND) (in crystalline γ-form), was acquired from Hawkins Pharmaceutical group (Minnesota, USA). The structural formulas are shown in Figure 1. Indomethacin was of United States Pharmacopeia grade, and tryptophan was reagent grade. All samples were used as received.

Figure 1. Chemical structures of the API (indomethacin) and amino acid (tryptophan) used in this model coamorphous system.

Milling. IND and TRP were milled alone and as a binary blend at a 1:1 molar ratio. The samples (1.2 g total for each sample) were ball milled in an oscillatory mill (Retsch ball mill PM 100, Haan, Germany) with a stainless steel grinding jar of 12 mL capacity and two stainless steel balls of 10 mm diameter at 650 rpm. Samples were milled for 360 min, with small amounts of sample (approximately 30 mg) removed for analysis at various time points. The zero minute sample was made by running the mill for 5 min without the use of grinding balls before adding the balls. At each sampling time the temperature of the powder was measured with a laser thermometer (mini IR thermometer 42500, Extech Instruments, MA, USA). Humidity was not controlled during milling. All samples were analyzed within 45 min of milling. XRPD. XRPD patterns of the samples were recorded with a PANanalytical X’Pert PROMPD system (PW3040/60, Philips, The Netherlands) using Cu Kα radiation with λ = 1.542 Å and a divergence slit of 1°. The powder was gently compressed in an aluminum sample holder. Samples were then scanned at 40 kV and 30 mA from 5° to 35° 2θ using a scanning speed of 0.1285 deg·min−1 and a step size of 0.0084°. The diffraction patterns were generated using X’Pert High Score version 2.2.0 (Philips, The Netherlands). Vibrational Spectroscopy. The samples were analyzed using various vibrational analytical techniques including Fourier-transform (FT) Raman spectroscopy, two lowfrequency Raman setups (with 785 and 830 nm excitation), and FT-infrared (FTIR) spectroscopy. Immediately after milling, approximately 10 mg of the sample was gently packed into aluminum divots and analyzed sequentially using the three Raman setups. Approximately another 10 mg was used for FTIR analysis. FT-Raman measurements were performed using a MultiRAM FT-Raman spectrometer (Bruker Optics, Ettlingen, Germany) equipped with a D418-T liquid nitrogen cooled Ge detector and a 2 W 1064 nm Nd:YAG laser. Spectra were collected using OPUS 6.5 (Bruker Optics, Ettlingen, Germany) B

DOI: 10.1021/acs.molpharmaceut.7b00803 Mol. Pharmaceutics XXXX, XXX, XXX−XXX


■

Article

RESULTS AND DISCUSSION XRPD. XRPD served as a reference with which to compare the Raman spectroscopic methods. Sharp diffraction peaks were visible in the diffractograms before milling (0 min) for both IND (including at 11.56, 16.93, 21.79, and 26.58° 2θ) and LTRP (including at 9.87, 14.85, and 19.84°2θ) revealing that both components were highly crystalline. IND was in the γ form (Cambridge Crystallographic Data Centre (CCDC) refcode INDMET01), while tryptophan was of the structure with CCDC refcode VIXQOK01 (Figure 2).

with a laser power of 150 mW, 64 coadded scans, a defocused laser spot (diameter of ∼1 mm), and a resolution of 2 cm−1. The 785 nm dispersive low-frequency Raman setup is a custom-built system based on a wavelength-stabilized 80 mW 785 nm laser module (Ondax, Inc., Monrovias, CA, USA). The laser line is initially filtered by use of two BragGrate reflective volume Bragg gratings (VBG) (OptiGrate Corp, Oviedo, FL) to remove amplified spontaneous emission. The sample is arranged in a 135° backscattering geometry relative to a collection lens (f/2.3). Collected light is passed through two VBGs (OptiGrate Corp. BragGrateTM 785 nm, OD3). The collimated, filtered light is then focused by a second f/2.3 lens onto a fiber optic cable, which is coupled to an LS 785 spectrograph (Princeton Instruments, Trenton, NJ, USA). A third VBG is used to further filter light imaged onto the entrance slit of the spectrograph. Detection is achieved using a thermoelectrically cooled PIXIS 100 BR CCD (Princeton Instruments, Trenton, NJ, USA). The spectral window −350 to 2080 cm−1 was collected with this setup with a resolution of 8 cm−1. This lower resolution across a wide window allowed for the simultaneous collection of spectra in both the low- and mid-frequency regions to directly compare how these two regions preformed for monitoring solid state changes induced by milling. Each spectrum was collected with 38 mW power with 1 s integration time × 300 coadditions giving a measurement time of ∼5 min. The 830 nm (200 mW) laser system is based on the SureBlockTM XLF-CLM THz-Raman system (Ondax Inc. CA, USA). The sample is arranged in a 180° backscattering geometry relative to a 10× microscope lens. This system is then coupled via a fiber optic cable to a SP2150i spectrograph and PIXIS 100 CCD (Princeton Instruments, Trenton, NJ, USA). The spectrograph was used in conjunction with a 1200 groove/mm blazed diffraction grating and a slit width of 30 μm. The spectra were acquired over the −70 to 940 cm−1 spectral window with a resolution of 4 cm−1. Each spectrum took approximately 10 min to acquire (600 coadditions with 1 s integration time or 40 coadditions with a 15 s integration time). FTIR measurements were performed on a purged Vertex70 Fourier Transform infrared spectrometer (Bruker Optics, Ettlingen, Germany) fitted with a GladiATR diamond ATR accessory (Pike Technologies, Madison, WI, USA). Spectra were acquired using OPUS 6.5 (Bruker Optics, Ettlingen, Germany). A background was acquired before each sample spectrum. Each spectrum was the result of 256 coadded scans over 50−4000 cm−1 at a 4 cm−1 spectral resolution. Multivariate Analysis. Principal component analysis (PCA) was performed on the vibrational spectra to help interpret subtle spectral variation between the milled samples at different time points. Linear baseline correction followed by a standard normal variate (SNV) transformation was performed on the spectra before PCA to remove baseline and intensity differences unrelated to the sample composition. PCA (NIPALS algorithm) was performed on the selected spectral regions after preprocessing and mean centering, leaving one sample out at a time for cross validation. The preprocessing and analysis was carried out using the Unscrambler X 10.3 (CAMO Software AS, Oslo, Norway). Heterospectral 2DCOS. Heterospectral 2DCOS was carried out using 2Dshige (Shigeaki Morita, Kwansei-Gakuin University, 2004−2005). The 830 nm low-frequency Raman data (5 to 250 cm−1) was correlated to the FT-Raman data (1500 to 1750 cm−1).

Figure 2. X-ray powder diffractograms of (a) IND, (b) TRP, and (c) 1:1 IND−TRP mixture as a function of milling time.

During milling, pure IND continued to exhibit crystallinity after 360 min, although peak broadening revealed that some long-range order was lost (Figure 2a). This peak broadening is likely associated with particle size reduction and crystal imperfections rather than exclusively the formation of amorphous domains. During milling the maximum temperature that the powder reached was 45 °C. This is similar to the Tg for pure amorphous IND, which may have affected amorphization. Amorphization was indeed slower and more limited compared with what was observed in work carried out by Jensen et al. on the same systems where IND was found to become amorphous at a faster rate than TRP when milled in a cooled environment.25 Pure TRP also exhibited peak broadening with peaks disappearing and the system becoming X-ray C


Article

Molecular Pharmaceutics amorphous after approximately 120 min (Figure 2b). TRP reached a maximum temperature of 50 °C during milling. Both compounds, particularly TRP, exhibited relative peak intensity changes within the first 5 min. This is due to initial particle size reduction during the first phase of milling decreasing preferred crystal orientation in the powdered material. The relative peak intensities then more closely mirrored those of the predicted powder diffraction patterns (with preferred orientation explicitly excluded). For the IND−TRP mixtures, peaks due to both IND and TRP were initially visible, though overall the IND peaks were more intense (Figure 2c). The overall intensity of the peak maxima decreased by a factor of approximately five within 5 min of milling, accompanied by peak broadening, consistent with increased solid-state disorder in the mixture. Preferred orientation also reduced for both materials. Resolved TRP peaks (for example, at 9.87 and 14.85° 2θ) had virtually disappeared within 10 min, while IND peaks were still visible until 90 min, at which point all peaks had disappeared to form an amorphous halo. The mixtures then remained X-ray amorphous for the remainder of milling. The maximum temperature reached during milling was 52 °C, despite these high temperatures, the comilled system became amorphous unlike IND milled alone. This is consistent with previous observation where the TRP acts as an antiplasticizer for the IND resulting in an increased Tg for the coamorphous system.25 The XRPD analysis reveals that the mixture loses order more easily and rapidly than the pure components alone. Increased rates and/or ease of amorphization with coamoprhous mixtures when compared to the pure components have previously been observed.26 This effect is likely to be facilitated by favorable cohesive intermolecular interactions (e.g., hydrogen bonding) between the drug and amino acid,27 though some degree of mechanical influence during the milling process may also be involved. The former mechanism is linked to direct coamorphization of the two components, while the latter would allow amorphization of the two components separately. It is difficult to draw further conclusions about the process and ultimate degree of molecular-level mixing of the two components (i.e., coamorphousness) from visual inspection of the XRPD data, and a Raman spectroscopy setup that is simultaneously capable of probing near range order, i.e., intermolecular interactions (mid-frequency Raman), as well as longer range order, i.e., crystallinity (low-frequency Raman), may answer this question. Vibrational Spectroscopy. Prior to milling, each sample exhibited Raman spectra with distinct bands in both the lowand mid-frequency regions. The Raman spectra gradually changed upon milling toward those that contained less intense peaks in both the low- and mid-frequency regions, with concurrent peak shifting in the mid-frequency region (Figure 3 and Supplementary Figure 1). In some heavily milled samples, the low-frequency regions simply showed a single broad band, consisting of the vibrational density of states (VDOS) in place of the previously distinct bands created by the crystalline sample (Figure 4 and Supplementary Figure 2). In the low-frequency range (10−250 cm−1), IND in its γcrystalline form featured bands at approximately 31, 46, 69, 95, 115, 149, and 195 cm−1. These bands are consistent with spectra collected by Hédoux et al. for γ-IND.20 The 785 nm low-frequency Raman data is presented in Figure 4, and the 830 nm low-frequency Raman data is given in Supplementary

Figure 3. FT-Raman spectra of (a) IND, (b) TRP, and (c) 1:1 IND− TRP mixture as a function of milling time.

Figure 2; please note that the 830 nm data appears to have outliers at the 90 and 120 min measurement points for the TRP sample when compared to other techniques, and as such these data are given in the Supporting Information. These outliers are possibly associated with subsampling as this setup consists of a ∼2 μm ⌀ diffraction limited sample spot, thus increasing the likelihood of subsampling. The low-frequency Raman spectra of the IND milled over time suggest that after 360 min of milling the indomethacin was still partially crystalline as the unique vibrational modes were still clearly observable. However, with milling duration there was some decreased intensity in the phonon modes and increased VDOS signal. The bands at 114 and 46 cm−1 collected on the 830 nm system appeared to be downshifted by ∼90 min; however, on fitting these bands with a mixed Gaussian−Lorentzian curve, it was observed that these shifts were less than 1 cm−1. It appears the visual inspection is misleading due to the underlying curve of the VDOS. However, such shifts are not inconsistent with particle size reduction due to phonon containment.28,29 This is also consistent with the XRPD results where the diffractograms exhibited some diminishing and broadening of features upon milling while still retaining the characteristic γ-form peaks, which can be characteristic of particle size reduction. There were no new or shifted peaks, confirming no transformation to an alternative crystalline polymorph. In the mid-frequency Raman spectra D


Article


visible at 120 min with the 785 nm setup. This peak was out of range for the FT-Raman setup, and consequently the sample appeared amorphous at the earlier time point of 120 min. Similar to IND, there were no new peaks or shifts that would indicate the appearance of a new polymorphic form. The midfrequency Raman spectra also showed changes upon increased milling duration with broadening and shifting of spectral features observed with both the FT-Raman and 785 nm systems (Figure 3b and Supplementary Figure 1b). Some particularly notable changes include the shifting of the band at 1558 to 1550 cm−1 upon milling for 360 min and the loss of many small sharp distinct bands from 1200 to 1500 cm−1. The band at 1558 cm−1 had shifted to approximately 1554 cm−1 at 90 min and 1551 cm−1 at 180 min with both setups. The peak appeared to be a single mode at all stages of milling, suggesting the system remained as a one phase system involving gradual structural change during milling. The IR spectra also contain similar changes to the Raman spectra with shifting and broadening of many spectral features. In particular, the loss of the bands at 1661 and 1582 cm−1 to form a broad feature centered around 1609 cm−1 and a peak shift from 495 to 532 cm−1 were evident with increased milling duration (Supplementary Figure 3b). With the comilled IND and TRP, the Raman spectra in the low- and mid-frequency ranges were dominated by features from IND since IND is a stronger Raman scatterer than TRP due to its relatively more delocalized structure. In the lowfrequency region, the crystalline bands associated with the compounds disappeared after approximately 60−90 min of milling (Figure 4c and Supplementary Figure 2c). There is some uncertainty in the interpretation of this due to the difficulty in distinguishing a clear feature at 68 cm −1 in the 830 nm data. It is hoped that by using multivariate techniques this visual inspection based discrepancy can be minimized. This is consistent with what was observed with XRPD. The midfrequency spectral region of the mixture shows peak broadening and shifting (Figure 3c and Supplementary Figure 1c). The simultaneous loss of the crystalline indomethacin feature at 1700 cm−1 and growth of the broad amorphous feature at 1680 cm−1 indicated a total loss of indomethacin crystallinity (and associated short-range order) after 90 min of milling. Careful inspection of the tryptophan modes at around 1558 cm−1 (for the crystalline form) and 1550 cm−1 (for the amorphous form) reveals that, unlike for the tryptophan milled alone, the two modes were simultaneously present at gradually changing relative intensities after milling for between 30 and 90 min. This is more obvious in the FT-Raman spectra. This duality combined with (a) the earlier appearance of the 1550 cm−1 mode at 30 min instead of 90 min for the pure tryptophan and (b) the earlier indomethacin spectral changes described above suggest that the system exists as multiple phases at this intermediate stage of milling. One of these phases is the coamorphous phase, with two others being the crystalline forms of the two components. Since the presence of the two components together promotes more rapid and complete amorphization than for either component alone, the presence of pure amorphous forms is likely to be absent or at least more limited than for the coamorphous form. This observation is in line with the results of a differential scanning calorimetry (DSC) study on the same mixture during milling in which only a single, albeit changing, Tg was observed at all milling times, consistent with the presence of a coamorphous form and the respective crystalline starting materials.25 It was suggested that

Figure 4. A 785 nm Raman (10−250 cm−1) of (a) IND, (b) TRP, and (c) 1:1 IND−TRP mixture as a function of milling time.

(450−1800 cm−1) from both the 785 nm Raman and FTRaman systems, band broadening indicating increased system disorder also occurred upon milling. In addition, peak shifts were also observed revealing a change in molecular confirmation and intermolecular interactions. A broad band ∼1680 cm−1, due to CO stretching of the benzoyl group, appeared and gradually increased (Figure 3a and Supplementary Figure 1a). In a fully amorphous sample, the band at 1700 cm−1 (which involves the same benzoyl moiety in the γcrystalline form), should completely disappear, leaving only the broad band at 1680 cm−1.18,30 The benzoyl CO group of each molecule interacts with the same group from a neighboring molecule in the γ-form, and a shift to a lower frequency upon loss of the crystalline form is due to the changed intermolecular environment.18 The unmilled TRP exhibited a series of modes at approximately 42, 77, 103, 127, and 166 cm−1 in the lowfrequency Raman systems (Figure 4b and Supplementary Figure 2b). Upon milling these modes gradually disappeared by 180 min to form a broad and diffuse signal typical of amorphous systems, and thus the system was considered amorphous at this point.31 The results are consistent with the XRPD results for this system with the samples being X-ray amorphous after 120 min of milling. The peak at 42 cm−1 was the last to disappear in the low-frequency region and was still E


Article


Table 1. Comparison of Time Points at which Samples Appeared Amorphous (Visual Inspection of the Spectra) and Spectra Stopped Changing (Based on PCA) IND experiment XRPD FT-Raman (450−1800 cm−1) FT-IR (450−1800 cm−1) 785 nm Raman (10−250 cm−1) 785 nm Raman (450−1800 cm−1) 830 nm Raman (10−250 cm−1)

visual inspection still partially 360 min) still partially 360 min) still partially 360 min) still partially 360 min) still partially 360 min) still partially 360 min)

TRP PCA

visual inspection

IND−TRP PCA

visual inspection

PCA

crystalline (at

90 min

120 min

120 min

90 min

90 min

crystalline (at

visually unclear 120 min

120 min

60 min

180 min

90 min based on features at 1700, 1680, 1558, and 1550 cm−1 60 min

30 min

crystalline (at

45−90 min 45−90 min 90 min

180 min

180 min

90 min

60 min

crystalline (at

90 min

180 min

45 min

90 min based on features at 1700, 1680, 1558, and 1550 cm−1 60−90 min

30 min

crystalline (at

visually unclear 180 min

crystalline (at

180 min

60 min

tryptophan changes alone and in combination when milled. The resulting scores plots were inspected to determine the approximate time point when the milling no longer induced further change (disorder) in the sample. In general, the XRPD and low-frequency Raman analyses (10−250 cm−1) required two PCs to describe both the predominant component present (IND or TRP) and the level of disorder. This is because with increased disorder the unique crystalline features from each component were lost and replaced with a broad amorphous halo, which was similar in its shape for both the individually milled compounds as well as the comilled system. The spectroscopic methods in the 450− 1800 cm−1 region required three PCs to describe this same variation; possibly because the disordered version of the comilled mixture was unique and could not be described with the same spectral variation occurring with the individually milled components. XRPD. For XRPD the first two PCs accounted for 75% of the overall sample variance, with the loadings plots revealing that PC1 was positively correlated with the crystalline IND signal. The negatively deviating baseline indicated it was also negatively correlated with the “halos” observed with the amorphous form. PC2 was positively correlated with crystalline TRP and negatively associated with disordered IND (Supplementary Figure 5). Inspection of these scores plots with the aid of each PC versus time plot suggests that, according to XRPD and PCA, the pure IND stopped structurally changing after 90 min of milling, TRP stopped changing after 120 min, and the IND−TRP mixture stopped changing after 90 min (Figure 5). This is consistent with the interpretation based on visual inspection alone for amorphization of TRP (120 min) and IND−TRP (90 min) samples. However, it is important to note that although the IND sample stopped changing, it was only partially disordered. The steady state reached for the IND with milling appears to be due to the disorder inducing process (milling) and recrystallization being in equilibrium. It is assumed that the temperature at which the milling process reached 45 °C facilitated this recrystallization as previous studies carried out in cooler conditions were able to create amorphous IND with milling.25 This information can be inferred not only from visual inspection of the diffractograms but also from the PC1 vs time scores plot. Both TRP and IND−TRP samples moved toward and cluster in negative PC1 and PC2 scores space with increased milling time, whereas the IND samples remained in positive PC1 space, associated with crystalline features of the diffractogram (Figure 5a).

the TRP was being dissolved in the amorphous IND and acting as an antiplasticizer to raise the Tg of the amorphous IND until the mixture was completely coamorphous.25 The ATR-IR spectra contains many changes, in particular, shifting and broadening of features in the 1750 to 1350 cm−1 region with samples appearing to be amorphous after 60 min of milling (Supplementary Figure 3c).28,29 A summary of the times at which the samples measured using XRPD and the various spectroscopic techniques appear to become amorphous (based on visual inspection of the spectra) is given in Table 1. In summary, based on visual inspection of the data, there is agreement between the low-frequency Raman and XRPD analysis as to when the samples are amorphous. Heterospectral 2DCOS was carried out on the comilled IND−TRP system to look at the relationship between changes observed with milling duration in the low-frequency and midfrequency Raman spectra. In the synchronous plot (Supplementary Figure 4a) the carbonyl stretch at 1699 cm−1 (IND) and to a lesser extent the 1620 cm−1 (IND and TRP) band were positively correlated with the features present at 31, 46, 69, and 98 cm−1 and negatively correlated with the broad VDOS signal in the low-frequency Raman spectra. The features at 1680 (amorphous IND), 1551 (TRP), and broad signal around 1585 cm−1 are positively correlated with the broad VSOS signal and negatively correlated with the features at 31, 69, and 98 cm−1 in the low-frequency Raman data. The associated asynchronous plot was much weaker than that of the synchronous plot (z-max of 0.26 versus 1.84) implying the asynchronous relationship is weaker than that of the synchronous. This plot suggests that the loss of the bands at 1699, 1620, 1589, and 1579 cm−1 in the mid-frequency region occurs before the growth of the VDOS in the low-frequency data. Similarly, the loss of the bands at 46, 69, and 98 cm−1 occur before the loss of the 1699, 1620, 1589, and 1579 cm−1 features, indicating earlier changes to the crystalline IND signals in the low-frequency Raman data. The growing of the VDOS, particularly around 15 cm−1, also occurs before the growth of the amorphous IND peak at 1680 cm−1, possibly highlighting the earlier sensitivity of changes induced by milling in the lowfrequency Raman data. However, further work needs to be done on this before drawing definitive conclusions. Multivariate Data Analysis. PCA was carried out on the various sets of data collected from XRPD and the multiple spectroscopic methods and spectral regions (i) to detect more subtle spectral changes based on whole spectral analysis and (ii) to compare the observed kinetic profiles of indomethacin and F


Article


Figure 5. PC scores values versus time for IND, TRP, and IND−TRP milled samples analyzed with XRPD: PC1 (upper) and PC2 (lower). Figure 6. PC scores values versus time for IND, TRP, and IND−TRP milled samples analyzed with 785 nm Raman setup over the 10−250 cm−1 spectral region: PC1 (left) and PC2 (right).

785 nm Raman. The 785 nm dispersive system was capable of collecting Raman spectra over the region −350 to +2080 cm−1 with the low-frequency region accessible to within 7 cm−1 of the laser line. This setup has the advantage of allowing the mid- and very low-frequency regions to be directly compared with one another without confounding experimental sources of variation. Thus, a detailed analysis of the multivariate data analysis with this setup is presented below. Low-Frequency Region (10−250 cm−1). Based on the PCA scores plot (Supplementary Figure 6a), PC1 separated signals associated with TRP structural changes during milling, while both PC1 and PC2 described the IND changes. This interpretation is supported by the loadings plots (Supplementary Figure 6b) in which the vibrational modes for TRP at around 42, 77, 103, 127, and 166 cm−1 are observed as depressions in the PC1 loadings plot. Disappearance of these modes during milling resulted in progressively more positive PC1 score values for the TRP samples. For IND, the bands observed at 31, 46, 69, 95, and 149 cm−1, among others, are all present as peaks in the PC2 loadings. The weakening of these signals during milling resulted in a decrease of the PC2 score values over time. The simultaneous increase in PC1 score values for IND over time is attributed to positive PC1 loadings consisting of a broad halo (VDOS) from the amorphous IND− TRP system with crystalline TRP signals subtracted; the decrease in intensity of the IND features is contributing to movement into neutral PC1 space. The comilled IND−TRP spectra appeared at points intermediate to those of the pure components but moved to more positive PC1 and negative PC2 values than those observed for either of the components alone. This is at least partly due to the loss of all distinct modes (in contrast to the IND alone), but may also reflect a distinct coamorphous spectral profile that is not merely the sum of those of the pure amorphous components (i.e., contains different intermolecular bonding and molecular conformations). As with the XRPD analysis, the PC scores versus time plots were used to determine the time scale of spectral changes and time to maximum disorder (Figure 6). The IND sample stopped changing after around 180 min of milling based on the

PC2 scores (and 90 min based on PC1 scores), suggesting that this was when the maximum level of disorder was reached. As earlier noted, the sample was not completely amorphous at either of these or subsequent time points, with the bands remaining at the end of milling (as well as in the XRPD data). This is somewhat evident from the PC2 values, which remained positive for the duration of milling (Figure 6). TRP stopped changing after between 60 and 120 min of milling based on the PC1 loadings (the 90 min sample was an outlier and excluded from consideration). The comilled IND−TRP system stopped exhibiting systematic spectral changes after approximately 30 to 60 min of milling. This is consistent with the absence of distinct spectral modes in the spectra at 60 min. As observed with visual inspection of the same data, PCA provided evidence that the mixture became amorphous more rapidly and easily than either of the pure components alone, in particular indomethacin. Mid-Frequency Region (450−1800 cm−1). The 785 nm spectra exhibited an emissive baseline for the IND containing samples. After the linear baseline correction, there was still variation associated with the curvature of the baseline, which was incorporated in the PCA. The first three PCs accounted for 97% of the explained variance. PC1 separated TRP from IND spectral features and did not provide useful information on crystallinity changes. This is to be expected, as the largest spectral differences in this region are associated with chemical, rather than physical differences. In contrast, PC2 was negatively correlated with crystallinity for TRP and to a lesser extent IND. PC3 also captured solid state changes, although this was different for IND and TRP: PC3 was positively associated with IND crystallinity but negatively associated with TRP crystallinity (Supplementary Figure 7). Inspection of PCs 2 and 3 versus time (Supplementary Figure 8) was used to determine the time point at which solid state changes ceased. IND alone stopped changing after around 90 min of milling, while the TRP spectra remained constant after around 180 min of milling. The IND−TRP comilled system appeared to stop changing after approximately 30 min of milling. This is earlier G


Article

Molecular Pharmaceutics than what was interpreted based on the visual inspection of specific peaks, with which a time to amorphization of 90 min was determined. This discrepancy is due to the noise across the spectral region subject to PCA, cumulatively overwhelming the individual peak changes studied with crystallinity change (specifically at 1700, 1680, 1558, and 1550 cm−1) after 30 min. It is possible that the lower spectral resolution and smaller sampling volume also contribute to this discrepancy. This frailty in the analysis highlights a further advantage of using the lowfrequency region for crystallinity analysis and for which no such discrepancy was observed: spectral differences associated with the solid state structure, rather than mere chemical composition predominate, in the low-frequency region. FT-Raman. The FT-Raman spectra in the low- and midfrequency regions were also subject to PCA analyses (Supplementary Figures 9 and 10). The time for the spectra to stop changing for the pure components and the mixture are summarized in Table 1, with maximum disorder being reached for IND, TRP, and the mixture after 45 to 90, 120, and 60 min, respectively. This technique did not suffer the same degree of discrepancy as the 785 nm setup for the PCA mid-frequency region analysis of the mixtures (though a difference of 30 min remained). This is likely to be due to the signal-to-noise ratio and the sample volume for the FT-Raman setup being higher than for the 785 nm setup. 830 nm Low-Frequency Raman. PCA of the 830 nm lowfrequency Raman data gave almost identical separation patterns to the 785 nm low-frequency Raman data (Supplementary Figures 11 and 12). PC1 (64% variance) predominantly separated based on TRP crystallinity (negatively correlated), and PC2 (32% variance) was positively correlated with IND crystallinity. Inspection of the scores plot suggests IND, TRP, and IND−TRP stopped changing after 45, 180, and 60 min of milling, respectively. FT-IR. PCA of the FTIR spectra gives similar separation patterns to that of the FT-Raman spectra with PC1 (79% variance) separating indomethacin from tryptophan signals, PC2 (12% variance) separating based on disorder of the TRP, and PC3 (6% variance) separating based on disorder of the IND (Supplementary Figure 13). In contrast to the FT-Raman results, the IR spectra gives stronger signals associated with the TRP compared with the IND. This is consistent with what is expected with IR and Raman selection rules and the associated molecular structures. PCs 2 and 3 describe the increase in disorder in the milled systems (Supplementary Figure 14). The times for the spectra to stop changing for the pure components and the mixture are summarized in Table 1, with maximum disorder being reached for IND, TRP, and the mixture after 45 to 90, 180, and 30 min, respectively. Direct Comparison of Techniques for Indomethacin and Tryptophan Mixture. PCA of the IND−TRP comilled sample was also carried out independently from the other samples for each technique and spectral region to determine the end point for changes due to milling and directly compare the different methods. The first PC for each of these methods was then dominated by differences in order/disorder of the samples rather than sample composition. The details for each individual technique’s PCA are given in the Supporting Information (Supplementary Figures 15 to 20). The scores for PC1 for each technique was then normalized to have a minimum of 0 and a maximum of 1 and directly compared to one another versus time (Figure 7).

Figure 7. Normalized PC1 value for amorphization of IND−TRP system with milling time.

Based on this analysis, the path to disorder follows a similar kinetic profile for all the techniques. The time to maximum disorder is 60 min for the FT-Raman, 830 nm setup, and XRPD data, while it appears to be slightly faster, at 30 min, for the 785 nm setup (both spectral regions). These differences in apparent times to the formation of the amorphous system could be associated with the different resolution of the systems used, with the higher resolution data resulting in a longer time to maximum disorder due to smaller variations in peak shape being detected.

■

CONCLUSIONS In this work, low-frequency Raman setups were compared with the more established mid-frequency Raman analysis for probing amorphous form creation. XRPD also served as a reference technique. The API, indomethacin, and the amino acid, tryptophan, were milled separately, and the induced disorder was investigated as a function of milling time by both direct visual inspection and PCA of the data. Unlike with the midfrequency Raman analyses, the disappearance of all distinct phonon modes in the low-frequency data made it is possible, like with XRPD, to directly determine when the systems became amorphous. The analyses also revealed that the mixture lost order faster and/or more completely than the individual components alone, which is likely to have been facilitated by favorable intermolecular interactions between the two components. PCA of whole spectral regions rather than individual peaks allowed the detected time and kinetic path to maximum disorder to be rapidly determined and compared among the analysis methods. Overall, low-frequency Raman spectroscopy has several favorable features when monitoring (co-)amorphous form creation: direct detection of complete loss of crystallinity (unlike mid-frequency Raman), a more intense Raman signal that is less prone to photoluminescence interference than the mid-frequency Raman region (a common problem with APIs), potential to combine with mid-frequency Raman data for more detailed chemical and intermolecular bonding analysis, and finally, potentially more versatile and mobile setups for in and online analysis than has been the case thus far for X-ray based analyses. The recent commercialization and availability of low-frequency Raman spectroscopy is likely to lead a rapid expansion of pharmaceutical applications in the near future. H


Article


■

(12) Aaltonen, J.; Alleso, M.; Mirza, S.; Koradia, V.; Gordon, K. C.; Rantanen, J. Solid form screening - A review. Eur. J. Pharm. Biopharm. 2009, 71, 23−37. (13) Rantanen, J. Process analytical technology applications of Raman spectroscopy. J. Pharm. Pharmacol. 2007, 59, 171−177. (14) De Beer, T. R. M.; Bodson, C.; Dejaegher, B.; Walczak, B.; Vercruysse, P.; Burggraeve, A.; Lemos, A.; Delattre, L.; Heyden, Y. V.; Remon, J. P.; Vervaet, C.; Baeyens, W. R. G. Raman spectroscopy as a process analytical technology (PAT) tool for the in-line monitoring and understanding of a powder blending process. J. Pharm. Biomed. Anal. 2008, 48, 772−779. (15) Nanubolu, J. B.; Burley, J. C. Investigating the Recrystallization Behavior of Amorphous Paracetamol by Variable Temperature Raman Studies and Surface Raman Mapping. Mol. Pharmaceutics 2012, 9, 1544−1558. (16) Aaltonen, J.; Heinänen, P.; Peltonen, L.; Kortejärvi, H.; Tanninen, V. P.; Christiansen, L.; Hirvonen, J.; Yliruusi, J.; Rantanen, J. In situ measurement of solvent-mediated phase transformations during dissolution testing. J. Pharm. Sci. 2006, 95, 2730−2737. (17) Savolainen, M.; Kogermann, K.; Heinz, A.; Aaltonen, J.; Peltonen, L.; Strachan, C. J.; Yliruusi, J. Improved insight into the dissolution behavior of amorphous drugs by in situ solid-state analysis. Eur. J. Pharm. Sci. 2008, 34, S27−S27. (18) Strachan, C. J.; Rades, T.; Gordon, K. C. A theoretical and spectroscopic study of gamma-crystalline and amorphous indometacin. J. Pharm. Pharmacol. 2007, 59, 261−269. (19) Roy, S.; Chamberlin, B.; Matzger, A. J. Polymorph discrimination using low wavenumber Raman spectroscopy. Org. Process Res. Dev. 2013, 17, 976−980. (20) Hedoux, A.; Paccou, L.; Guinet, Y.; Willart, J.-F.; Descamps, M. Using the low-frequency Raman spectroscopy to analyze the crystallization of amorphous indomethacin. Eur. J. Pharm. Sci. 2009, 38, 156−164. (21) Hedoux, A.; Decroix, A.-A.; Guinet, Y.; Paccou, L.; Derollez, P.; Descamps, M. Low- and high-frequency Raman investigations on caffeine: polymorphism, disorder and phase transformation. J. Phys. Chem. B 2011, 115, 5746−5753. (22) Larkin, P. J.; Dabros, M.; Sarsfield, B.; Chan, E.; Carriere, J. T.; Smith, B. C. Polymorph characterization of active pharmaceutical ingredients (APIs) using low-frequency Raman spectroscopy. Appl. Spectrosc. 2014, 68, 758−776. (23) Hubert, S.; Briancon, S.; Hedoux, A.; Guinet, Y.; Paccou, L.; Fessi, H.; Puel, F. Process induced transformations during tablet manufacturing: Phase transition analysis of caffeine using DSC and low frequency micro-Raman spectroscopy. Int. J. Pharm. 2011, 420, 76−83. (24) Mah, P. T.; Fraser, S. J.; Reish, M. E.; Rades, T.; Gordon, K. C.; Strachan, C. J. Use of low-frequency Raman spectroscopy and chemometrics for the quantification of crystallinity in amorphous griseofulvin tablets. Vib. Spectrosc. 2015, 77, 10−16. (25) Jensen, K. T.; Larsen, F. H.; Cornett, C.; Lobmann, K.; Grohganz, H.; Rades, T. Formation Mechanism of Coamorphous Drug-Amino Acid Mixtures. Mol. Pharmaceutics 2015, 12, 2484−2492. (26) Lobmann, K.; Grohganz, H.; Laitinen, R.; Strachan, C.; Rades, T. Amino acids as co-amorphous stabilizers for poorly water soluble drugs - Part 1: Preparation, stability and dissolution enhancement. Eur. J. Pharm. Biopharm. 2013, 85, 873−881. (27) Lobmann, K.; Laitinen, R.; Strachan, C.; Rades, T.; Grohganz, H. Amino acids as co-amorphous stabilizers for poorly water-soluble drugs - Part 2: Molecular interactions. Eur. J. Pharm. Biopharm. 2013, 85, 882−888. (28) Ivanda, M.; Furic, K.; Music, S.; Ristic, M.; Gotic, M.; Ristic, D.; Tonejc, A. M.; Djerdj, I.; Mattarelli, M.; Montagna, M.; Rossi, F.; Ferrari, M.; Chiasera, A.; Jestin, Y.; Righini, G. C.; Kiefer, W.; Goncalves, R. R. Low wavenumber Raman scattering of nanoparticles and nanocomposite materials. J. Raman Spectrosc. 2007, 38, 647−659. (29) Arora, A. K.; Rajalakshmi, M.; Ravindran, T. R.; Sivasubramanian, V. Raman spectroscopy of optical phonon confine-

ASSOCIATED CONTENT

* Supporting Information S

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.molpharmaceut.7b00803. Spectra and PCA models (PDF)

■

AUTHOR INFORMATION

Corresponding Authors

*E-mail: clare.strachan@helsinki.fi *E-mail: [email protected]. ORCID

Holger Grohganz: 0000-0002-0482-1397 Keith C. Gordon: 0000-0003-2833-6166 Notes

The authors declare no competing financial interest.

■

ABBREVIATIONS ATR, attenuated total reflectance; FTIR, Fourier transform infrared; IND, indomethacin; PCA, principle component analysis; SNV, standard normal variate; TRP, tryptophan; VBG, volume Bragg gratings; VDOS, vibrational density of states; XRPD, X-ray powder diffraction

■

REFERENCES

(1) Nagapudi, K.; Jona, J. Amorphous active pharmaceutical ingredients in preclinical studies: Preparation, characterization, and formulation. Curr. Bioact. Compd. 2008, 4, 213−224. (2) Williams, H. D.; Trevaskis, N. L.; Charman, S. A.; Shanker, R. M.; Charman, W. N.; Pouton, C. W.; Porter, C. J. H. Strategies to address low drug solubility in discovery and development. Pharmacol. Rev. 2013, 65, 315−499. (3) Laitinen, R.; Löbmann, K.; Strachan, C. J.; Grohganz, H.; Rades, T. Emerging trends in the stabilization of amorphous drugs. Int. J. Pharm. 2012, 453, 65−79. (4) Paudel, A.; Nies, E.; Van den Mooter, G. Relating hydrogenhonding interactions with the phase behavior of Naproxen/PVP K 25 solid dispersions: evaluation of solution-cast and quench-cooled films. Mol. Pharmaceutics 2012, 9, 3301−3317. (5) Baird, J. A.; Taylor, L. S. Evaluation of amorphous solid dispersion properties using thermal analysis techniques. Adv. Drug Delivery Rev. 2012, 64, 396−421. (6) Löbmann, K.; Laitinen, R.; Grohganz, H.; Gordon, K. C.; Strachan, C.; Rades, T. Coamorphous drug systems: Enhanced physical stability and dissolution rate of indomethacin and naproxen. Mol. Pharmaceutics 2011, 8, 1919−1928. (7) Löbmann, K.; Strachan, C.; Grohganz, H.; Rades, T.; Korhonen, O.; Laitinen, R. Co-amorphous simvastatin and glipizide combinations show improved physical stability without evidence of intermolecular interactions. Eur. J. Pharm. Biopharm. 2012, 81, 159−169. (8) Karmwar, P.; Graeser, K.; Gordon, K. C.; Strachan, C. J.; Rades, T. Investigation of properties and recrystallisation behaviour of amorphous indomethacin samples prepared by different methods. Int. J. Pharm. 2011, 417, 94−100. (9) Mah, P. T.; Laaksonen, T.; Rades, T.; Aaltonen, J.; Peltonen, L.; Strachan, C. J. Unravelling the relationship between degree of disorder and the dissolution behavior of milled glibenclamide. Mol. Pharmaceutics 2014, 11, 234−242. (10) Chang, C. F.; Okajima, H.; Hamaguchi, H.; Shigeto, S. Imaging molecular crystal polymorphs and their polycrystalline microstructures in situ by ultralow-frequency Raman spectroscopy. Chem. Commun. 2014, 50, 12973−12976. (11) Aaltonen, J.; Strachan, C. J.; Pollanen, K.; Yliruusi, J.; Rantanen, J. Hyphenated spectroscopy as a polymorph screening tool. J. Pharm. Biomed. Anal. 2007, 44, 477−483. I


Article

Molecular Pharmaceutics ment in nanostructured materials. J. Raman Spectrosc. 2007, 38, 604− 617. (30) Surwase, S. A.; Boetker, J. P.; Saville, D.; Boyd, B. J.; Gordon, K. C.; Peltonen, L.; Strachan, C. J. Indomethacin: New Polymorphs of an Old Drug. Mol. Pharmaceutics 2013, 10, 4472−4480. (31) Parrott, E. P. J.; Zeitler, J. A. Terahertz Time-Domain and LowFrequency Raman Spectroscopy of Organic Materials. Appl. Spectrosc. 2015, 69, 1−25.

J


Probing Pharmaceutical Mixtures during Milling - ACS Publications

Recommend Documents