Phase Separation in Coamorphous Systems: in Silico Prediction

Eight different drug–drug mixtures were studied to identify the miscibility; terfenadine with paracetamol, indomethacin, and ASA; indomethacin with ...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/molecularpharmaceutics

Phase Separation in Coamorphous Systems: in Silico Prediction and the Experimental Challenge of Detection Katja Pajula,*,† Lieke Wittoek,‡ Vesa-Pekka Lehto,§ Jarkko Ketolainen,† and Ossi Korhonen† †

School of Pharmacy, University of Eastern Finland, POB 1627, FI-70211 Kuopio, Finland Department of Pharmaceutical Analysis, Faculty of Pharmaceutical Sciences, Ghent University, Harelbekestraat 72, 9000 Ghent, Belgium § Department of Applied Physics, University of Eastern Finland, POB 1627, FI-70211 Kuopio, Finland ‡

S Supporting Information *

ABSTRACT: Combinatorial chemistry has enabled the production of very potent drugs that might otherwise suffer from poor solubility and low oral bioavailability. One approach to increase solubility is to make the drug amorphous, which leads to problems associated with drug stability. To improve stability, one option is to molecularly disperse the drug in a matrix. However, the primary reason for the failed stabilization with this approach is phase separation, which has been carefully studied in polymeric systems. Nevertheless, the amorphous−amorphous phase separation in coamorphous small molecule mixtures has not yet been reported. The goal of the present study was to experimentally detect the amorphous−amorphous phase separation between two small molecules. A modified in silico method for predicting miscibility by the Flory−Huggins interaction parameter is presented, where conformational variations of the studied molecules were taken into account. A series of drug−drug mixtures, with different mixture ratios, were analyzed by conventional differential scanning calorimetry (DSCconv) to detect possible amorphous− amorphous phase separations. The phase separation of coamorphous drug−drug mixtures was also monitored by temperature modulated DSC (MDSC) and Fourier transform infrared (FT-IR) imaging at temperatures above Tg for prolonged time periods. Amorphous−amorphous phase separation was not detected with DSCconv, probably due to the slow kinetics of phase separation. However, the melting of the separated and subsequently crystallized phases was detected by MDSC. Furthermore, FT-IR imaging was able to detect the separation of the two amorphous phases, which demonstrates the ability of this method to detect small molecule phase separations. KEYWORDS: amorphous, coamorphous mixture, phase separation, long-term stability, Fourier transform infrared (FTIR) imaging

1. INTRODUCTION 1,2

parameters (e.g., the Flory−Huggins interaction parameters). Solubility parameters may also be defined by group contribution theory.16−19 The miscible system is generally considered stable. The preceding or simultaneous step of recrystallization in immiscible two component mixes is the phase separation,4 which happens with some kinetics over time. It must be emphasized that thermodynamic miscibility is the equilibrium property that changes as a function of both temperature and composition. Without knowing the temperature dependency of miscibility, one can certainly obtain a miscible system at high temperature but the phases may still separate at lower temperatures. This can happen especially in melt−quench and hot−melt extrusion applications. Also, it has been shown that compounds having a low crystallization tendency (i.e., easily produced pure amorphous form) with “intermediate” miscibility

2,3

The high apparent solubility, dissolution rate, and thermodynamic instability2,4−8 are typical characteristics of molecular amorphicity. It is generally accepted that a storage temperature of 50 °C below the glass transition temperature (Tg) will minimize the risk for recrystallization because of the resulting low molecular mobility.2,9 However, molecular mobility still occurs in the glassy state, even though the mobility is much slower when compared with the rubbery state. Because of the higher viscosity, recrystallization may still occur during the expected shelf life of the product.9−11 Thus, amorphous drugs must be formulated with excipients that thermodynamically stabilize, or kinetically diminish, the recrystallization tendency. Ideally, the excipient selection should be science-based, fast, and cost-effective.12 The miscibility or “amorphous solubility” of a solid is the main thermodynamic parameter for this process and, subsequently, the critical measure that defines the thermodynamic state of a solid mixture at equilibrium.4 The miscibility between two compounds can be evaluated by either calculated13−15 or experimentally determined16,17 solubility © XXXX American Chemical Society

Received: November 26, 2013 Revised: March 28, 2014 Accepted: May 13, 2014

A

dx.doi.org/10.1021/mp400712m | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

several algorithms (steepest descent, ABNR, and quasiNewton), and thus, most likely attains the global minimum structure. The convergence limits during geometry optimization for energy change, maximum force, and maximum displacement between optimization steps were 2 × 10−5 kcal/mol, 0.001 kcal/mol/Å, and 1 × 10−5 Å, respectively. The maximum number of geometry optimization steps was 5000. After the geometry optimization the molecule with the final energy value was saved to the results file. On the basis of these settings, the final results file contained 100 conformations for each studied molecule. Energy values were plotted in increasing order. If the energies grew stepwise, one conformation was selected from each step. Alternatively, if all calculated conformations had a different energy value, then the 10 lowest energy conformations were included as starting structures for subsequent miscibility calculations. 2.2.2. Calculating the Flory−Huggins Interaction Parameter. Eight different drug−drug mixtures were studied to identify the miscibility; terfenadine with paracetamol, indomethacin, and ASA; indomethacin with paracetamol, ASA, and glibenclamide; and finally, paracetamol with glibenclamide and ASA (Table 1). The miscibility between the studied pairs was determined by calculating all selected conformations systematically with each other (Materials Studio 5.5, Blends toolbox, Accelrys Software Inc.). If both molecules have 10 individual conformations, then each possible combination yields 100 pairs of molecules and, consequently, 100 miscibility values. Details of the calculation routine are presented elsewhere.15 The calculations started from the determination of binding energies, and the energy of mixing, Emix, was obtained from eq 1:

(i.e., a Flory−Huggins interaction parameter close to zero or slightly positive) can apparently form miscible, single-phase systems in the short term, especially when mixing is performed at high temperature and subsequently low viscosity. Subsequent cooling to a low temperature increases the viscosity and decreases the kinetics of phase separation, as an apparent single-phase system is formed. Such systems are thermodynamically unstable, in regard to miscibility. Hence, even though a miscibility prediction is based on thermodynamic mixing, the thermodynamics and kinetics cannot be totally separated, if the primary goal is to achieve a satisfactory shelf life for amorphous drugs. In a previous article, we described a systematic approach to the selection of the small molecule additives for stabilization of amorphous drug compounds with the varying crystallization tendencies.15 That selection method was based on in silico predictions of the Flory−Huggins interaction parameter for the miscibility of a two-component system. The starting point for calculations was the global minimum energy conformation of the compounds. This computer assisted modeling method was able to predict miscibility with high precision, except for mixtures where at least one of the components had lower crystallization tendency than the other. These systems remained amorphous in the short term, despite the predicted immiscibility. However, this did not exclude the possibility of an immiscible system, as the amorphous−amorphous separation could have occurred without detection. Phase separation in polymeric systems has been widely reviewed,4,16,20−27 but phase separation in coamorphous small molecule mixtures has not been reported. For this reason, an analytical method for predicting the miscibility of a solid, twocomponent system is presented herein, where the conformational variability of the compounds is taken into account. The first goal was to experimentally verify the predicted immiscibility with long-term experiments, and the second goal was to detect the kinetically hindered amorphous− amorphous phase separation with mixtures containing two small molecular compounds with low crystallization tendency.

ΔEmix =

1 [Z bs(E bs)T + Zsb(Esb)T − Z bb(E bb)T 2 − Zss(Ess)T ]

(1)

where Z is the coordination number, E is the binding energy, and s and b denote screen and base, respectively. The temperature-dependent Flory−Huggins interaction parameter, χAB, was then derived from eq 2:

2. MATERIALS AND METHODS 2.1. Materials. Terfenadine, indomethacin, and acetylsalicylic acid (ASA) were obtained from Sigma-Aldrich. Paracetamol was obtained from Oriola (Espoo, Finland), and glibenclamide was obtained from Hangzhou Dayangchem Co., Ltd. (Hangzhou, China). 2.2. Predicting the Miscibility. 2.2.1. Selecting the Conformations for Miscibility Calculations. The structures of drug molecules were constructed with Materials Studio 5.5 (Accelrys Software Inc.). The conformational space of the studied molecules was examined by the Quench protocol in Material Studio (Materials Studio 5.5, Forcite module, Accelrys Software Inc.). Calculations were performed by COMPASS force field and force field assigned charges. The atom-based method, with no truncation, was used to calculate electrostatic charges and van der Waals energies. In Quench protocol, molecular dynamic simulation was performed with the constant-temperature and constant-volume ensemble (NVT), also referred as canonical ensemble. The temperature control method was the Hoover variation of the Nosè-thermostat with Q-ratio of 0.01. The temperature (2000 K) was set high to ensure the free rotation for all bonds. A total of 1,000,000 steps were simulated with a step size of 1 fs. After every 10,000 steps, the existing conformation was quenched and geometry was optimized with Smart algorithm, which is a compilation of

χAB =

ΔEmix RT

(2)

where R and T are the gas constant and temperature, respectively. Finally, the Gibbs free energy change of mixing was determined as a function of temperature and composition from eq 3: ΔGmix = RT (nA ln ØA + nB ln ØB + χAB nA ØB)

(3)

where ni and Øi are the respective number of moles and volume fractions of components A and B. This calculation used the same energy settings as above, i.e., COMPASS force field and force field assigned charges. The electrostatic and van der Waals energies were determined by using an atom-based method with no truncation cutoff limits. In the miscibility calculation, the number of pair configurations for which the binding energy was calculated was 10,000,000. The width of the bins that were used to generate the energy distribution was 0.02 kcal/mol. The coordination number was calculated from 100,000 cluster samples. The number of iterations used to build each cluster was 20, and the miscibility was evaluated at 293 K. Probability distribution and cumulative distribution plots of the Flory−Huggins interaction B

dx.doi.org/10.1021/mp400712m | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

melting temperature of the coamorphous system (50/50 molar ratio) Tm50:50, according to eq 4,

parameters were used to evaluate the miscibility of the studied compounds. 2.3. Methods for Detecting Phase Separation. Four different methods were used to detect phase separation; conventional differential scanning calorimetry (DSC), modulated differential scanning calorimetry (MDSC), Fourier transform infrared imaging (FT-IRi), and polarized light microscopy. 2.3.1. Conventional DSC. Conventional DSC was used to analyze the glass transition temperatures (Tg) and melting points (Tm) of the pure compounds and their crystallization tendency as a reference. DSC was also used to study the phase separation in a series of varying compositions, and to prepare the amorphous binary systems for subsequent stability studies. 2.3.1.1. Glass Transition Temperatures and Melting Points of Pure Compounds and the Crystallization Tendency. The compounds were analyzed by a DSC (Mettler Toledo DSC 823e, Switzerland) coupled to a refrigerated cooling accessory (Mettler Toledo, METT-FT900 Julabo, Switzerland) and an autosampler (Mettler Toledo, TSO801RO, Sample Robot, Switzerland) to determine the Tg and Tm of the pure compounds, which were used reference points to determine the Tg of subsequent drug−drug mixtures. Samples were kept at 25 °C for 1 min, and then heated from room temperature to 5 °C above the melting point with a heating rate of 10 °C/min. The samples were held at the melt temperature for 5 min, then quenched under an uncontrolled rate to −50 °C, and held there for 15 min. Finally, the samples were heated again to 10 °C above the melting temperature at a heating rate of 10 °C/min. The DSC cell was purged with nitrogen (50 mL/min). Temperature and heat flow were calibrated using reference standards (i.e., indium, lead, zinc, and highly purified water). Samples were weighed with a high precision microbalance (Sartorius SE2, Sartorius AG, Germany), and typical weights of the samples were within a few milligrams and measured to an accuracy of 0.1 μg. Analyses were made in sealed 40 μL aluminum pans (Mettler Toledo, Switzerland) having a pierced lid. Each measurement was performed in triplicate. The results were analyzed with STARe software (Mettler Toledo, Switzerland). The glass transition temperatures were determined as midpoint values and melting points as onset values. If crystallization was not detected during the temperature program, then the compound was categorized as noncrystallizing, whereas moderately crystallizing compounds crystallized during the second heating. If a compound crystallized during the quench step, it was categorized as fast-crystallizing. 2.3.1.2. Analyzing the Series of Binary Mixtures for Miscibility and Phase Separation. The molar ratio of the studied drug−drug mixtures were 0:100, 10:90, 30:70, 50:50, 70:30, 90:10, and 100:0, which were prepared by mixing the compounds and grinding them with an agate mortar and pestle. Each mixture was studied with the same DSC method described in section 2.3.1.1. The 50:50 mixtures were also analyzed to determine the melting temperature, which was used to reduce the temperature scale between Tg and Tm. 2.3.1.3. Preparing the Coamorphous Samples for Stability Study. The coamorphous drug−drug samples (50/50 molar ratio) for stability studies were prepared as described in section 2.3.1.1, with the exception that the second heating was increased to the storage temperature. The samples were then placed in desiccators containing silica, stored at elevated temperature above Tg, and analyzed after 2 and 5 weeks. The storage temperature (Ts) for each mixture was determined by the Tg (as described in section 2.3.2) of the system and the

Ts − Tg Tm50:50 − Tg

= 0.3 (4)

The Tm50:50 of coamorphous mixtures were analyzed by conventional DSC (as described in section 2.3.1.2). The samples were stored at 0.3 in the reduced temperature scale. This approach was taken to equalize the temperature scales between the different binary systems, and all samples were stored at the same relative percentage position between Tg and Tm and analyzed in duplicate. 2.3.2. Analyzing the Stored Samples with Temperature Modulated DSC (MDSC). Temperature modulated DSC was used due to its high sensitivity in detecting weak and partially overlapping transitions to analyze the stored samples and Tg of the coamorphous drug−drug mixtures. Analyses were done at a slower heating rate when compared with conventional DSC methods used. The method is described in section 2.3.1.1, with the exception of stochastic temperature modulation (TOPEM), which was used during the second heating. The underlying heating rate for the second heating was 2 °C/min with the pulse height of 0.5 K. The switching time varied between 30 and 60 s. The measured Tg was used for the reduced temperature scale (eq 4) TOPEM temperature modulation was also used to analyze drug−drug mixtures after the stability study. Initially, samples were held for 1 min at 25 °C, quenched to −50 °C for 15 min, and finally heated to 10 °C above the melting point by the TOPEM temperature modulation. 2.4. Long-Term Stability Testing with Fourier Transform Infrared Imaging (FT-IRi) and Polarizing Light Microscopy. Amorphous−amorphous phase separation was monitored from the coamorphous drug−drug mixtures by FTIRi in transmission mode (Spectrum Spotlight, PerkinElmer, Shelton, CO). The samples were prepared on top of a zinc selenide IR-glass by melting a small amount of physical mixture (50/50 molar ratio) and then placing another IR-glass on top of the sample. A region of 0.1 mm2, with 63 points (equal spacing of 50 μm with the array of 3 in y-axis by 21 in x-axis), was measured. The pure amorphous compounds were prepared with the same way as drug−drug mixtures to identify the characteristic peaks for pure compounds with no overlapping, the only exception being ASA due to its fast crystallization rate as a pure compound. Hence, the reference spectra for ASA were defined from the crystalline powder (cf. ref 15). The wavenumbers from 2000 to 720 cm−1 were used, with a resolution of 4 cm−1. The spot size was 6.25 μm, and the number of scans per measured point was two, as these produced good spectra with low noise. No spectral pretreatment methods were used. The interpretation of the FT-IR spectra and image generation was done with Spotlight Software (version 1.0, PerkinElmer, Inc.). The intensity change of the characteristic peaks in the spectra of drug−drug mixtures was used to detect the phase separation in the measured area. The analyzing was started from the spectra measured at the time point of 14 days to identify the greatest change, and then the time point was identified, where the change was first observable. Before IR-imaging, the samples were analyzed by polarizing light microscopy (Nikon LV100D, Japan) for the evidence of amorphicity. Thus, only amorphous samples were measured with IRi. The samples were stored at elevated temperatures, as reported in Table 3, and analyzed with both FT-IRi and C

dx.doi.org/10.1021/mp400712m | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

illustrates the interaction parameters of paracetamol−ASA over a wide range of values, and thus, a single interaction parameter was not able to describe the miscibility behavior of paracetamol−ASA. The interaction parameter between the global minimum energy conformations of paracetamol−ASA (standard approach in miscibility predictions) was roughly −5, which indicates that paracetamol and ASA are miscible with each other (Figure 3 and Table 1), while the highest probability value

microscopy after 1, 2, 3, 7, 10, and 14 days of storage. In addition, two mixtures out of eight were stored for 21 weeks to monitor the initiation of crystallization by polarized light microscopy.

3. RESULTS 3.1. Miscibilities Predicted by a Modified Calculation Method of the Flory−Huggins Interaction Parameter. The selection of conformations for miscibility calculations were based on the following criteria: (1) if there were less than ten conformations with different total energy values, then each conformation was included in calculation (Figure 1), and (2) if

Figure 3. Distribution of the interaction parameters for the paracetamol−ASA mixture. The interaction parameter between the global minimum energy conformations is indicated by the red dot. Figure 1. Energy values for the four different molecular conformations of paracetamol, which were used for miscibility calculations.

of 1 indicates only a slight immiscibility between paracetamol and ASA (Figure 3). Further, the cumulative distribution plot shows that 58% of the interaction parameter values were below zero and 42% above zero, which delineate the miscibility and immiscibility, respectively (Figure 4). These results demonstrate that the calculated mixture was in the vicinity of the miscible/immiscible boundary limit and that paracetamol−ASA was only a partially miscible mixture. For the indomethacin−paracetamol mixture, the distribution plot indicated that a majority of the calculated interaction parameters were near a value around 5 (Figure 5). This indicates an immiscibility between the studied compounds. Also, according to the cumulative distribution plot, approximately 85% of the calculated interaction parameters were positive, and the global minimum energy conformation was 3, thus supporting the preceding interpretation that compounds were immiscible (Figures 5 and 6). In the case of indomethacin−paracetamol, the standard global minimum energy conformation interaction parameter would have provided a reliable prediction, but not any indication of variability of interaction parameters. The remaining mixtures studied were analyzed according to the interpretation stated above, and these results are presented in Table 1. For paracetamol−glibenclamide mixture, the interaction parameter with the highest probability (χa in Table 1) and an interaction parameter between the global minimum energy conformations (χb in Table 1) indicated miscibility, as the values were approximately 0 and −7, respectively. However, as over 60% of the calculated interaction parameters were positive (χ% > 0 in Table 1), this mixture was classified as immiscible. The rest of the mixtures studied, except paracetamol−ASA, were classified as immiscible with each other. 3.2. Glass Transition Temperatures, Melting Points, and Crystallization Tendencies of Pure Compounds. The Tg, Tm, and crystallization tendencies of pure compounds were analyzed as a reference for studies on the phase separation behavior of amorphous drug−drug mixtures (Table 2 and Supporting Information). All the compounds presented in this

every conformation had a different energy value, ten conformations with the lowest total energies were calculated (Figure 2) (Supporting Information). Results are presented from predictions at 293 K and a 50/50 molar ratio.

Figure 2. Ten of the lowest energy conformations (in red) of terfenadine that were used for miscibility calculations.

Three different measures of the interaction parameter were used in the evaluation of miscibility for the studied compounds; (1) the interaction parameter of the highest probability, (2) the interaction parameter between the global minimum energy conformations, and (3) the percentage of interaction parameters that were above zero (immiscible) from the cumulative distribution plot of interaction parameters (Supporting Information). In this way, a more realistic description of the behavior of miscibility can be achieved by taking into account the conformational variability in the evaluation of miscibility predictions. For the paracetamol−ASA mixture, four distinct peaks with interaction parameter values of approximately −7, −4, 1, and 4 are presented in the probability plot (Figure 3). This example D

dx.doi.org/10.1021/mp400712m | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

Table 1. Calculated Interaction Parameters (χ) That Were Used in the Miscibility Prediction and the Time (Days) the CoAmorphous Mixtures Remained Amorphous, Despite Poor Mixing Behavior (n = 3); in General, Positive and Negative χ Values Are Indicative of System Immiscibility and Miscibility, Respectively MW (g/mol)

N:oconfd

additive

MW (g/mol)

N:oconfd

χa

χb

χ% > 0

tcrystc mean (days)

±SD

terfenadine

471

100

indomethacine

357

100

paracetamol

151

4

paracetamol indomethacine ASA paracetamol ASA glibenclamide glibenclamide ASA

151 357 180 151 180 494 494 180

4 100 6 4 6 100 100 6

3.9 11.4 6.7 4.7 7.9 4.6 −0.2 0.6

0.9 4.7 6.2 2.6 7.7 6.7 −6.5 −4.7

60.1 85.9 76.3 84.8 86.5 67.5 61.1 42.4

2.3 nde 6.3 1.3 0.3 5.7 32.3 2.7

0.6 nde 3.1 0.6 0.6 2.3 18.5 3.8

drug

a The highest probability χ. bχ parameter from the global minimum conformations. ctcryst = crystallization time measured by polarized light microscope. dN:oconf = number of conformations. end = not detected.

study were moderately- or noncrystallizing (Table 2). Hence, all mixtures were expected to be amorphous after preparation. The question remained if these were phase separated systems or apparently miscible systems that will eventually phase separate over time, and how to detect the amorphous− amorphous phase separation. In our previous paper, we utilized crystallization as an indirect measure of immiscibility or phase separation of binary mixtures.15 This approach worked well with compounds that possessed high crystallization rate as pure component. However, if the pure compound had a low crystallization tendency, or was amorphous, then crystallization could not serve as an indirect measure of immiscibility or phase separation of the binary mixture. 3.3. Miscibility Indicated by the Change in Tg as a Function of Content. It is generally thought that detection of a single Tg for a mixture is indicative of the miscibility of the components, whereas two Tg values indicate the system to be partly miscible. If the detected Tg values are close to the corresponding values of the pure components and if these Tg values are independent of the composition, the system is regarded as immiscible. Disadvantages of above-mentioned classification are fairly low sensitivity with low content systems (the onset of phase separation) and low resolution when the Tg values of pure components are close to each other. The general features of miscibility phase diagrams are that the extreme compositions tend to be immiscible, which is utilized in the present study. All compositions were initially single phase systems having only one Tg (Figures 7−9 and Supporting Information). Thus, the systems were apparently miscible, or the second Tg was below the DSC detection limit. Also, Tg values of some compounds were quite close to each other, and two Tg values were not detected for this reason. If immiscible systems were apparently miscible, the kinetics of phase separation was too slow to be detected just after their preparation, even if the added content was very low. Hence, a single Tg detected right after preparation only described the state of the mixing, not the thermodynamic miscibility of a system. The immiscibility leading to phase separation has to be studied experimentally as a function of time, if one or both compounds are slowly crystallizing as pure compound. 3.4. Detection of Phase Separation in 50:50 Mixtures with DSC after Five Week of Storage at Elevated Temperature. The Tg and Tm values were used to determine the elevated storage temperatures for studied drug−drug mixtures (Table 3). The second heating in the sample preparation was made up to the determined storage temperature

Figure 4. Cumulative distribution plot of interaction parameters for the paracetamol−ASA mixture.

Figure 5. Distribution of the interaction parameter for the indomethacin−paracetamol mixture. The interaction parameter of global minimum energy conformations is indicated by the red dot.

Figure 6. Cumulative distribution plot of interaction parameters for the indomethacin−paracetamol mixture. E

dx.doi.org/10.1021/mp400712m | Mol. Pharmaceutics XXXX, XXX, XXX−XXX

Molecular Pharmaceutics

Article

Table 2. Drug−Drug Mixtures (50:50 in Molar Ratio) with Glass Transition Temperatures (Tg), Melting Points (Tm), and Crystallization Tendencies of Pure Compounds Analyzed by Conventional DSC with the Heating Rate of 10 °C/min (n = 3) Tm (°C)

±SD

Tg (°C)

±SD

additive

Tm (°C)

±SD

Tg (°C)

±SD

terfenadinea

148.9

0.2

60.0