In-Line Monitoring of Hydrate Formation during Wet Granulation Using Raman Spectroscopy

HAKAN WIKSTROM, PATRICK J. MARSAC, LYNNE S. TAYLOR Industrial and Physical Pharmacy, School of Pharmacy, Purdue University, West Lafayette, Indiana 47907

Received 5 April 2004; revised 30 July 2004; accepted 13 September 2004 Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.20241

ABSTRACT: Process-induced transformations are very important to control during pharmaceutical manufacturing because they may change the properties of the active pharmaceutical ingredient in the drug product, compromising therapeutic efcacy. One process that may facilitate a process-induced transformation is high-shear wet granulation. In this study, the feasibility of Raman spectroscopy for in-line monitoring of the transformation of theophylline anhydrous to theophylline monohydrate during high-shear wet granulation has been evaluated. The midpoint of conversion occurred 3 min after the binder solution was added. The effects of several processing parameters were also examined, including mixing speed and monohydrate seeding. Mixing speed had the greatest effect on the transformation, where an increase in mixing speed shortened the onset time and increased the rate of transformation. In contrast, seeding with monohydrate or changing the way in which the binder was incorporated into the granules did not affect the transformation prole. The transformation kinetics observed during wet granulation were compared with those generated by a simple model describing the solvent-mediated transformation of theophylline in solution. In conclusion, these studies show that Raman spectroscopy can be used for in-line monitoring of solid-state transformations during wet granulation. In addition, for this particular compound, a simple solvent-mediated transformation model has been shown to be useful for estimating the time scale for hydrate formation during high-shear wet granulation.
2004 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 94:209219, 2005

Keywords:

transformation; hydrate; granulation; Raman spectroscopy

INTRODUCTION
Process-induced transformations (PITs) occur as a result of mechanical or thermal stress imposed upon a system during processing or after exposure to solvent. PITs are well documented and include production of amorphous regions, crystallization of amorphous material, polymorphic transformations, change in size and shape of materials,

Correspondence to: Lynne S. Taylor (Telephone: 765-4966614; Fax: 765-494-6545; E-mail: ltaylor@pharmacy.purdue.edu)
Journal of Pharmaceutical Sciences, Vol. 94, 209219 (2005) 2004 Wiley-Liss, Inc. and the American Pharmacists Association

dehydration of crystalline hydrates, and hydration of anhydrous crystals.1 PITs are important because they may alter the properties of the nal dosage form such as the solubility, dissolution rate, hygroscopicity, stability, solid-state reactivity, and ultimately bioavailability. One process that may facilitate a PIT is wet granulation in which the blended components of a formulation are agitated as water is added leading to granule formation and growth due to mobileliquid bonding between primary particles. The quantity of water needed to achieve good granules varies with formulation and processing conditions, but can be up to 50% (w/w) of the dry powders. Although techniques such as differential scanning calorimetry, X-ray powder diffraction, solid-state
209

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

210

WIKSTROM, MARSAC, AND TAYLOR

nuclear magnetic resonance, and microscopy can be helpful in identifying PITs off-line, an in-line test, that is, in situ monitoring in real-time, that minimizes disturbance of the process and eliminates artifacts associated with sample preparation is obviously advantageous to improve process understanding and control. To date, it seems that the only technique that has been used to monitor phase transformations in-line during wet granulation is X-ray powder diffraction,2 although both Raman and near infrared (NIR) spectroscopy have been used at-line for this purpose.3 Raman spectroscopy is emerging as a useful technique for pharmaceutical process monitoring, particularly during the synthesis and production of drug substance. Svensson et al.4 demonstrated that Raman spectroscopy combined with chemometric analysis could be used for reaction monitoring. Raman spectroscopy has also been used to monitor the solvent-mediated polymorphic transformation of progesterone5 and in the identication and quantitation of three polymorphic forms of a developmental drug compound during a slurry conversion.6 Given the well-documented ability of Raman spectroscopy to discriminate between solid-state forms, even in the presence of excipients,79 and the availability of commercial process spectrometers with suitable sampling congurations, it is of interest to investigate whether this technique can be used for in-line monitoring of phase transformations during high-shear wet granulation. In this study, theophylline was selected as the model compound and the kinetics of transformation were monitored using in-line Raman spectroscopy. The mechanism of theophylline transformation to the hydrate in an aqueous environment has been extensively investigated.10,11 The monohydrate form is the most stable form below 608C and heterogeneously nucleates from the surface of the anhydrate in water. In this investigation, we were interested in probing the transformation during wet granulation in the presence of excipients. The effects of mixing speed, changes to the active pharmaceutical ingredient (API), including seeding and ball-milling, and changes to binder solution were examined. Finally, we attempted to predict the timescale for transformation during wet granulation using a simple model for solventmediated transformations, which assumes NoyesWhitney dissolution,12 a screw dislocationmediated growth-rate equation,10 and a simple mass balance.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

EXPERIMENTAL
Materials Theophylline anhydrous (AT) was purchased from Rhodia (Cranbury, NJ) and placed in an oven at 1008C for at least 1 h before use to ensure that no monohydrate was present and used immediately after cooling. The surface area of AT was measured using a Micromeritics ASAP 2010 BET (Micromeritics Instrument Company, Norcross, GA) with nitrogen gas. Eight hundred milligrams of AT was placed in a round-bottom ask and the material was degassed overnight at 1008C. Analyses were performed in triplicate. Theophylline monohydrate (MT) was supplied by Mallinckrodt Chemical Works (St. Louis, MO) and placed in a desiccator with 100% relative humidity for a week before use to ensure that no anhydrate was present. Avicel-PH-101 microcrystalline cellulose (MCC) was obtained from FMC Corporation (Newark, DE). Mannitol was obtained from Ruger Chemical Company (Irvington, NJ) and screened through a 20-mesh screen to reduce agglomeration before used. Polyvinyl pyrrolidone K-29/32 (PVP) was obtained from ISP Technologies, Inc. (Wayne, NJ). The binder solution was prepared by slowly adding 100 g of PVP to 800 mL of double-distilled water. Slurry Experiments Slurry conversion experiments were conducted in triplicate using a jacketed vessel and a Neslab RTE-111 circulated water bath (Neslab Instruments, Inc., Newington, NH). Five grams of AT was placed in 40 mL of doubly distilled water maintained at 25 18C and the slurry was agitated with a magnetic stir bar. Wet Granulation Dry material was weighed and placed in a Diosna P 1/6 high-shear mixer-granulator (Dierks & Sohne GmbH, Osnabruck, Germany) equipped with a 2-L stainless steel bowl. All granulations were performed according to Table 1, unless otherwise noted. The binder solution was sprayed onto the mix using a Masterex Quick Load model 7021-24 pump (Cole-Parmer Instrument Co., Vernon Hills, IL) and the amount of binder solution added was determined gravimetrically using a Mettler PC 8000 balance (Mettler Toledo, Inc., Hightstown, NJ).

IN-LINE MONITORING OF HYDRATE FORMATION

211

Table 1. Typical Batch Components and Operating Conditions AT MCC Mannitol PVP (added in water solution) Water Mixing speed Chopper speed Dry mixing time Binder solution addition time Wet massing time 90 g 105 g 105 g 14.5 g 115 g 100 rpm 1200 rpm 2 min 0.6 min 10 min

Calibration Calibration samples were prepared in triplicate by geometrically mixing AT with MT in 10% (mol/ mol) increments with a total of 4 g of material per sample. In addition, 5 and 15% (mol/mol) levels of both forms were also included to better describe the extremes of the calibration range. To minimize particle size differences, AT was prepared from MT by placing it in an oven at 1008C for several days to ensure complete transformation. To minimize the effect of sub-sampling, an automatic sampling device was constructed consisting of an electrical motor that rotated the sample vial. The at-faced immersion probe was tted with a blade and inserted into the sample bed. An additional blade was inserted into the vial, positioned so that it cleaned the sides of the vial from sticking material. The blade tted to the immersion probe was positioned so that it moved material from the bottom of the vial. With this setup, an adequate calibration curve was obtained from triplicate sample preparations, each measured three times. Software HoloGRAMS software (version 3.0; Kaiser Optical Systems) was used to control the Raman spectrometer. Spec32 software (version 4.0; Control Development) was used to control the NIR spectrometer. Excel (build 9.0.2720; Microsoft Corporation, Seattle, WA) was used for calibration calculations and graph plotting. SIMCA-P (version 10.0.4; Umetrics AB, Umea, Sweden) was used for partial least squares (PLS) data analysis. Raman spectra from HoloGRAMS software were transferred to Excel via SIMCA as SPC les (Thermo Galactic, Salem, NH) and NIR spectra from the Spec32 software were imported to SIMCA as ASCII text les. Sigma Plot (version 8.02; SPSS, Inc., Chicago, IL) was used for curve tting, and Polymath (version 5.1; CACHE Corporation, Austin, TX) was used for the modeling calculations.

Raman Spectroscopy Raman spectra were collected using an RXN1785 Raman spectrometer (Kaiser Optical Systems, Inc., Ann Arbor, MI) equipped with a 1/400 MultiRxn stainless steel immersion probe with a at sapphire window. A 10400 mW diode laser at 784.8 nm was used for excitation and the power at the sample was measured to be around 100 mW using a LaserCheck power meter (Coherent, Inc., Auburn, CA). Each spectrum was composed of two scans with a minimum integration time of 5 s per scan. Spectra were obtained every 1530 s over the duration of the experiments. For wet granulation experiments, the immersion probe was placed in the mixing bowl just above the impeller and angled toward the movement of the bed to reduce the risk of sample adhesion. For some experiments, a round-tipped probe from Matrix Solutions, Inc. (Seattle, WA), designed by Brian J. Marquardt, was used to further minimize the risk of sample adhesion to the probe. Repeat runs were performed using both probes to verify that the probe design did not inuence the results.

NIR Spectroscopy In-line NIR experiments were performed using a NIR-250L-1.7T2 NIR spectrometer (Control Development, Inc., South Bend, IN) equipped with a 35-W tungsten halogen light source with a gold reector mounted on a lamp xture with a focusing lens mounted in front of an optical ber. The spectrometer was calibrated against a 50-mm Halon/Albrillon white reectance reference (Control Development) before use. Spectra were continuously accumulated over the range of 1100 2200 nm using an integration time of about 0.025 s. Only data from 1100 to 1850 nm was used because of saturation of the detector at longer wavelengths.

RESULTS
Calibration To determine the amount of theophylline transformed at each time point during the wet granulation experiments, a calibration curve was generated from powder blends of known ratios of
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

212

WIKSTROM, MARSAC, AND TAYLOR

AT and MT. As can be seen from Figure 1, AT has distinct peaks at 1664 and 1707 cm1 and MT has a distinct peak at 1686 cm1, all related to carbonyl vibrations in the theophylline molecule. Because the excipients used in these studies have little or no Raman response in this region, these peaks were chosen to construct the calibration curve. Because there is slight overlap between these peaks, a calibration model for overlapping peaks was used to determine the ratio of AT and MT in the samples.13 The observed-versus-predicted plot was tted to a linear regression line with the equation y 0.9937x 5.431 and exhibits sufcient linearity with a correlation coefcient (R2) value of 0.988, thus indicating that Raman spectroscopy can discriminate between the anhydrous and hydrate forms of theophylline over the entire concentration range. The NIR calibration was calculated using a PLS regression model and the NIR spectra of AT and MT can be seen in Figure 2. The data matrix was pretreated using a standard normal variate transformation,14 mean centered, and t to a two principal component PLS regression model. The R2 and Q2 values were 0.994 and 0.994, respectively, indicating that the model had a good t and predictability. The root mean square error of prediction was determined to be 4.4 by predicting known samples at 20, 40, 60, and 80% MT (mol/mol). Wet Granulation It has been well documented that theophylline undergoes a solvent-mediated phase transfor-

Figure 2. NIR spectra of AT and MT from 1100 to 1850 nm.

Figure 1. Raman spectra of anhydrous (AT) and monohydrate (MT) theophylline from 1500 to 1750 cm1.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

mation from the anhydrate to the monohydrate during wet granulation.11 Because off-line analyses have several disadvantages, including sampling issues and time delays between sampling and analysis, we were interested in assessing the feasibility of in-line Raman and NIR spectroscopy as methods to monitor the transformation in real time during the granulation experiment. NIR spectroscopy has been interfaced with many processing operations; however, the reported use of Raman spectroscopy for pharmaceutical formulation unit operations seems to be limited to at-line analyses. Before interfacing the sampling devices with the process, the mass of binder solution necessary to produce acceptable granules was determined. The mean volume diameter of the resultant granules was 350 mm, with 80% of the particles being within 100 mm of the average particle size. The Carrs index was determined to be 18% and attrition was negligible. When the sampling devices were present in the mixing bowl, there was no visually discernible change in the quality of the granules, that is, the devices did not appear to cause noticeable disruption of the granulation process. Figures 3 and 4 show examples of Raman and NIR spectra collected at various time intervals during wet granulation and the zero time point corresponds to the beginning of binder addition. From Figure 3, it is apparent that peaks characteristic of AT at 1664 and 1707 cm1 disappear and the peak corresponding to MT at 1686 cm1

IN-LINE MONITORING OF HYDRATE FORMATION

213

Figure 3. Waterfall plot of Raman spectra collected during granulation showing the transformation of AT with characteristic peaks at 1664 and 1707 cm1 to MT with a characteristic peak at 1686 cm1.

Figure 5. Effect of mixing speed on kinetics of transformation of AT to MT monitored with Raman spectroscopy during wet granulation. Binder addition was initiated at t 0 and data before this point were acquired during dry mixing.

appears during wet massing indicating that it is possible to follow the transformation with Raman spectroscopy. Furthermore, because there is no signicant interference from either water or excipients in this spectral region, the changes in peak height ratios could be converted to percentage transformed using calibration data. The resulting time prole of hydrate formation is shown in Figure 5. In contrast, the main differences in the NIR spectra between the two forms of theophylline arise because of the presence of the water molecule, which results in increased absorption at 1450 nm, as seen in Figure 2. Thus, it is not possible to follow the transformation kinetics with NIR spectroscopy because of the drastic changes water addition has to these regions, even if the

wavelength regions used for the calibration model were selected so that the inuence from the water signal was minimal. However, it is possible to gain information about binder addition. The NIR data show changes in the water content whereas Raman data provide information about the kinetics of theophylline hydrate formation. Because we were primarily interested in investigating transformation kinetics, only Raman spectroscopy was used for more extensive investigations. Having established that Raman spectroscopy could be used to monitor the transformation, three batches of identical composition were prepared using the same granulation conditions (see Table 1) in order to evaluate the repeatability of this methodology. For this particular formulation subject to these process variables, the time taken for 50% of the theophylline to transform was 3 min after initiating binder solution addition with a relative standard deviation of 6.0%. This level of repeatability provides an indication that it should be possible to study the effect of process variables on the transformation. Inuence of Mixing Speed Differences in mixing speed are, among other parameters, likely to affect the rate of water distribution, wetting, and the shear forces experienced by particles. To investigate the inuence of this parameter on the transformation kinetics, granulations were performed using different mixing speeds. As a way of comparing the different granulation conditions, data obtained at mixing speeds of 100, 200, and 400 rpm (corresponding to
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

Figure 4. Waterfall plot of NIR spectra collected during a granulation run highlighting the dominating effect of the water peak at 1450 nm.

214

WIKSTROM, MARSAC, AND TAYLOR

tip speeds of 0.9, 1.8, and 3.6 m/s, respectively) were t to the empirical eq. (1). This equation was chosen because it t the data well and allowed us to compare granulations with different processing conditions. CMT min max min k 1 t=mid 1

Effect of API Properties A granulation was prepared in which 5% (mol/ mol) of monohydrate seeds were substituted for AT powder. The rate of transformation was similar to that obtained without seeding. The effect of increasing the concentration of theophylline in the formulation was also investigated. The theophylline loading was increased by 50% from 30 to 45% (w/w), keeping the ratio of mannitol to microcrystalline cellulose constant at 1:1 (w/w). The amount of water was decreased slightly to better suit this formulation. No differences between the transformation rate in the formulation consisting of 45% (w/w) API and the formulation consisting of 30% (w/w) API could be seen. Finally, ball-milled theophylline was substituted for unprocessed theophylline to investigate the effect of initial surface area and surface properties on the rate of transformation. Although successful in-line monitoring of granulation was completely impossible because of excessive sticking to the probe, periodic sampling indicated that the transformation was completed in <3 min compared with only 50% being transformed in that time during identical process condition with unprocessed AT. Effect of Changes to Binder Addition The effect of adding the binder as a dry powder instead of as an aqueous solution was also investigated. No signicant changes occurred with respect to the rate of transformation; however, adding binder as a dry powder signicantly changed the granules properties.

where CMT is the fraction of monohydrate, min is the initial level of CMT (i.e., 0), max is the nal level of CMT (i.e., 1), mid is the time when 50% has been transformed, k is related to the rate of transformation, and t is time in minutes. The mean midpoint (n 3) was reached at 2.9, 2.1, and 1.2 min after the beginning of binder solution addition for the mixing speeds of 100, 200, and 400 rpm, respectively (see Table 2). These results indicate that mixing speed inuences the transformation kinetics, and that for the 100 400 rpm range of mixing speeds, transformation time decreases with increasing mixing speed. A plot of the transformation proles for different mixing speeds is shown in Figure 5. In addition to the experiments described above, granulations at lower mixing speeds (30 and 50 rpm, i.e., tip speeds of 0.27 and 0.45 m/s) were also performed. Very large granules were produced, which interfered with data collection. Offline measurements were made after the granulation was complete and it seems likely that the distribution of water was inhomogeneous because the extent of transformation measured at different locations in the granulation bowl varied signicantly. At higher mixing speeds (600 rpm or at tip speeds >5.4 m/s), mixing was so rigorous that it increased the temperature of the granules, resulting in an increase in transformation time, probably due to the decrease in supersaturation at higher temperatures. In addition, at 600 rpm, the batch quickly became overgranulated causing material to stick to the probe, which prevented representative sampling.

DISCUSSION
Interfacing the Raman spectrometer to the wet granulation process proved to be relatively straightforward for the lab-scale equipment used in these studies yielding spectra of sufcient quality in a reasonable time interval such that

Table 2. Effect of Mixing Speed on Transformation Midpoint, Condence Interval (CI) About the Midpoint, and Slope of AT to MT Conversion Prole Mixing Speed 100 rpm 200 rpm 400 rpm
a

Midpoint 2.93 min 2.06 min 1.18 min

Lower CIa 2.50 min 1.55 min 0.86 min

Upper CIa 3.37 min 2.57 min 1.50 min

Slope 5.7 7.7 11.3

95% CI.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

IN-LINE MONITORING OF HYDRATE FORMATION

215

the kinetics of transformation could be monitored. However, it is important to consider that, when using immersion optics, in order to obtain spectra that are representative of the bulk, material in close proximity to the probe must constantly be replenished. In our experiments, this was achieved through placing the probe in the vicinity of the impeller so that material was sheared past the bottom of the probe. This approach was successful for a at-faced probe only for certain granulations; in several instances, sticking to the probe was observed, suggesting that probe placement is quite critical. Switching to a ball probe (a probe tipped with a hemispherical window) reduced the tendency for material to stick to the probe although for some experiments, for example, when using milled theophylline or lowering the rotation speed, sticking was again observed to be a problem. These observations suggest that noncontact optics would be the preferable method of interfacing Raman spectroscopy to the wet granulation process. The potential advantages of noninvasive sampling are, among others, minimum disturbance of the process, reduced contamination risk, and greater exibility when changing equipment. Transformation kinetics of theophylline is rapid and complete conversion to the hydrate occurs during a relatively short granulation cycle. It is pertinent to consider that the solubility of AT is approximately 12 mg/mL at 258C, hence at any time point, the total mass of theophylline that can be dissolved in the granulation is approximately 1.5 g or <2% of the total mass of theophylline present assuming that all the added water is available for dissolving the drug substance. This suggests that theophylline undergoes a solventmediated transformation during wet granulation. This is consistent with previous observations that aqueous suspensions of theophylline undergo solvent-mediated transformation.11 Solvent-mediated transformations have been extensively studied and modeled and it is known that three processes take place and control the kinetics: dissolution of the less stable phase, nucleation of the more stable phase, and growth of the more stable phase.15 Once the induction time has passed, the rate of transformation may be controlled by either the dissolution of the metastable phase or by the growth of the more stable phase or by a combination of both factors. The rate at which the steady-state concentration is approached and the steady-state concentration during the transformation both depend on the kinetics

of dissolution and growth as described by Cardew and Davey.15 Consider that growth of the stable phase will not begin until the concentration in solution has exceeded the solubility of the stable phase. Growth of the stable phase will then deplete the concentration in solution whereas dissolution of metastable phase replenishes the solution. The steady-state concentration for a dissolution ratelimited transformation will therefore be similar to the solubility of the stable form whereas the steady-state concentration for a growth ratelimited transformation will be similar to the solubility of the metastable form. The seeding experiments indicate that, for this particular compound, transformation to hydrate is not rate limited by nucleation to the new phase. This is consistent with results presented by Rodrguez-Hornedo et al.,11 who suggested that MT nucleates via an epitaxial mechanism on the surface of AT crystals. In addition, microscopic examination of the AT used in this study showed growth of the monohydrate phase on the surface of anhydrate crystals upon addition of water (data not shown) indicating that the surface acts as a heterogeneous nucleation site. Hence, the formation of hydrate will likely be dependent on the surface characteristics of the starting material. Prediction of Transformation Kinetics The hydration of theophylline during aqueous wet granulation is a solvent-mediated transformation that takes place in a complicated system; excipients and theophylline are mixed under high shear, binder solution is metered into the mixture, the powders are wetted, some materials (theophylline, mannitol) may dissolve into the water whereas others (microcrystalline cellulose) may absorb the water, and granule growth occurs. Meanwhile, the water provides a medium through which AT dissolves, MT nucleates on the surface of AT, and MT grows. Several models in the literature describe solvent-mediated transformations under wellcontrolled conditions. Cardew and Davey15 have developed a model that predicts the timescale of a solvent-mediated transformation based on the characteristic crystal size of the starting material and the nal crystal size of the transformed material. Using a population balance,16 Thompson and Dixon17 extended the model outlined by Cardew and Davey15 to include the crystal size distribution (CSD). Rodrguez-Hornedo and Wu10 modeled the conversion of AT to monohydrate in
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

216

WIKSTROM, MARSAC, AND TAYLOR

terms of the initial concentration in solution and the initial and nal CSDs of each solid phase. Clearly, the wet granulation environment is a much more complicated system and measuring the nal CSD and the solution phase concentration during wet granulation would be a difcult, if not impossible, task. Raman spectroscopy only provides information about the percentage of each solid phase. However, it is possible to predict the timescale of the transformation based on a few simple equations in order to identify whether a solvent-mediated transformation would pose a potential problem during wet granulation. For example, Davis et al.2 predicted the polymorphic transformation of ufenamic acid by calculating the solubility of each phase, the growth rate constant, assuming rst-order kinetics, and tting the dissolution rate constant to on-line X-ray data collected during wet granulation. Taking a similar approach, assuming Noyes-Whitney dissolution and second-order growth with respect to supersaturation,10 we have attempted to predict the rate of theophylline hydrate formation during wet granulation. This model does not take into account the induction period. The surface area available for dissolution and nucleation of MT was estimated by tting the model to slurry data using published rate constants and solubilities. Dissolution of AT AT is assumed to follow Noyes-Whitney dissolution and is expressed in terms of the degree of undersaturation according to eq. (2). dA Csol SA kd AA dt SA 2
Figure 6. Scanning electron microscopy pictures of AT crystals highlighting the presence of agglomerates. a) 400 magnication, b) 6000 magnication.

where kd is the dissolution rate constant, AA is the surface area of AT, Csol is the concentration of theophylline in solution, and SA is the solubility of AT. The dissolution rate constant was measured by de Smidt et al.18 to be 0.043 mg/s using a rotating disk method at 160 rpm. The intrinsic dissolution rate constant was calculated to be 0.0146 mg/ mm2 min by dividing the dissolution rate constant by the surface area of the rotating disk.19 The initial external surface area of AT was estimated to be 0.14 m2/g by tting the model to a slurry experiment and compares well with the surface area measured by nitrogen gas adsorption: 0.46 m2/g. The lower value can be explained by the fact that AT tends to form agglomerates, as shown in Figure 6 and it has been shown that
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

surface area for dissolution decreases with formation of agglomerates.20,21 Next, the solubility of the AT was taken to be 12.3 mg/mL at 258C as reported previously11 and the change in surface area with time was estimated from the change in mass of the dissolving phase, assuming constant crystal shape factors.22 Growth of MT It has been shown that the growth of MT onto MT seeds is proportional to the square of the supersaturation with a temperature-dependent growth

IN-LINE MONITORING OF HYDRATE FORMATION

217

rate constant.10 From these results, we estimate a growth rate constant of 16.3 mm/min at 258C. Assuming that this growth rate constant can be applied to our system and multiplying the growth rate constant by the density as measured by Ebisuzaki et al.,23 an intrinsic growth rate constant of 0.0246 mg/mm2/min is obtained. The solubility of MT was taken to be 5.96 mg/mL at 258C10 and the rate of monohydrate growth was modeled with eq. (3).   dM Csol SM 2 kG AM 3 dt SM where kG is the growth rate constant, AM is the surface area of MT, and SM is the solubility of MT. It was then assumed that AM remained constant throughout the process. This is a reasonable assumption because growth of MT is linear and nucleation can be assumed to occur as an initial burst implying constant surface area during growth.10 Mass Balance To complete the model, the total mass must be conserved as shown in eq. (4), such that the sum of the changes in mass in the solution phase, the anhydrous phase, and the monohydrate phase must be zero.   dCsol dA dM V 4 dt dt dt where V is the volume of binder solution. Modeling Wet Granulation A comparison of the model described in the previous section to the data collected during granulation is given in Figure 7. The observed transformation time for the granulation is well described by the model, particularly for mixing speeds of 200 and 400 rpm. This observation, in conjunction with the results of the different granulation experiments, provides insight into some of the factors controlling the transformation. First, although the water/theophylline ratio is much greater for the slurry compared with the granulation, the transformation kinetics are comparable, thus the quantity of water used in wet granulation is clearly not a limiting factor. Furthermore, any hydrodynamic differences between the slurry conversion experiments and the granulations seem to have a minimum effect on the
Figure 7. Comparison between modeled transformation prole of AT to MT and transformation prole after the onset of transformation monitored using Raman spectroscopy during wet granulation for various mixing speeds.

transformation. Additionally, seeding with monohydrate had no effect on the induction or transformation time, suggesting that nucleation is not the limiting factor (note that the amount of water added at the beginning of the process was not enough to dissolve more than a fraction of the seeds added). Although the origin of the induction period has not been elucidated (Fig. 5), likely it is associated with water distribution, wetting of theophylline, and the initial dissolution stage. It is also interesting to note that the excipients used in this study had no perceivable inuence on the transformation kinetics. The two factors that did exert an inuence were impeller speed and ballmilling. It is likely that both of these can inuence the growth of monohydrate. A higher impeller speed would be expected to provide more shear force, which in turn would lead to better water distribution. This effect was not as dramatic as ball-milling AT. For this sample, the transformation time was practically instantaneous. It is likely that milling the powder changed the surface properties and resulted in an increased number of high-energy sites (including the possible formation of amorphous regions) for nucleation of MT and an increased surface area available for dissolution.

CONCLUSIONS
In-line Raman spectroscopy was successfully interfaced to high-shear wet granulation of a formulation containing AT and used to monitor
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

218

WIKSTROM, MARSAC, AND TAYLOR

the solvent-mediated psuedopolymorphic conversion of the API. NIR spectroscopy could not be used to monitor the phase transformation in this system because of the large absorbance of bulk water, which hid spectral information relating to the transformation. Because the conversion occurred over a relatively short time period (<5 min), using an in situ method enabled proling of the transformation kinetics. The conversion rate to MT was found to be similar for high-shear wet granulation and a simple slurry system. The ability to follow the transformation allowed considerable insight into the critical parameters determining the conversion rate. Raman spectroscopy shows considerable potential as an advanced in-line monitoring technique particularly for the understanding and control of process-induced solid-state transitions, although advances in instrumentation need to be progressed to realize this potential.

4.

5.

6.

7.

8.

9.

ACKNOWLEDGMENTS
Dr. Jukka Rantanen and Francis E. Rhea are gratefully acknowledged for assistance with data analysis and experimental support. The authors are grateful to Mary A. Albrecht at SSCI, Inc. for obtaining the scanning electron microscopy pictures. Brian J. Marquardt and Matrix Solutions, Inc. are acknowledged for the design and use of the round-tip immersion probe used for some Raman experiments. Xiaomin Lui of Illinois Institute of Technology is thanked for helpful discussions on modeling. Finally, AstraZeneca R & D Molndal is acknowledged for nancial support and Kaiser Optical Systems, Inc. is acknowledged for assistance with instrumentation.

10.

11.

12.

13.

14.

REFERENCES
1. Morris KR, Griesser UJ, Eckhardt CJ, Stowell JG. 2001. Theoretical approaches to physical transformations of active pharmaceutical ingredients during manufacturing processes. Adv Drug Delivery Rev 48(1):91114. 2. Davis TD, Morris KR, Huang HP, Peck GE, Stowell JG, Eisenhauer BJ, Hilden JL, Gibson D, Byrn SR. 2003. In situ monitoring of wet granulation using online X-ray powder diffraction. Pharm Res 20(11): 18511857. 3. Rasanen E, Rantanen J, Jorgensen A, Karjalainen M, Paakkari T, Yliruusi J. 2001. Novel identica-

15.

16.

17.

18.

tion of pseudopolymorphic changes of theophylline during wet granulation using near infrared spectroscopy. J Pharm Sci 90(3):389396. Svensson O, Josefson M, Langkilde FW. 1999. Reaction monitoring using Raman spectroscopy and chemometrics. Chemom Intell Lab Syst 49(1): 4966. Wang F, Wachter JA, Antosz FJ, Berglund KA. 2000. An investigation of solvent mediated polymorphic transformation of progesterone using in situ Raman spectroscopy. Org Proc Res Dev 4(5):391395. Zhou G, Wang J, Ge Z, Sun Y. 2002. Ensuring robust polymorph isolation using in-situ Raman spectroscopy. Am Pharm Rev 5:7478. Davies MC, Binns JS, Melia CD, Hendra PJ, Bourgeois D, Church SP, Stephenson PJ. 1990. Ft Raman-spectroscopy of drugs in polymers. Int J Pharm 66(13):223232. Hendra PJ. 1996. Fourier transform Raman spectroscopy in pharmaceutical analysis and research. Am Lab 28(18):17. Taylor LS, Langkilde FW. 2000. Evaluation of solidstate forms present in tablets by Raman spectroscopy. J Pharm Sci 89(10):13421353. Rodrguez-Hornedo N, Wu HJ. 1991. Crystalgrowth kinetics of theophylline monohydrate. Pharm Res 8(5):643648. Rodrguez-Hornedo N, Lechuga-Ballesteros D, Wu HJ. 1992. Phase-transition and heterogeneous epitaxial nucleation of hydrated and anhydrous theophylline crystals. Int J Pharm 85(13):149 162. Noyes AA, Whitney WR. 1897. The rate of solution of solid substances in their own solution. J Am Chem Soc 19:930934. Taylor LS, Zogra G. 1998. The quantitative analysis of crystallinity using FT-Raman spectroscopy. Pharm Res 15(5):755761. Barnes RJ, Dhanoa MS, Lister SJ. 1989. Standard normal variate transformation and de-trending of near-infrared diffuse reectance spectra. Appl Spectrosc 43(5):772777. Cardew PT, Davey RJ. 1985. The kinetics of solvent-mediated phase-transformations. Proc R Soc London Ser A 398(1815):415428. Randolph AD, Larson MA. 1988. Theory of particulate processes analysis and techniques of continuous crystallization. 2nd ed. San Diego: Academic Press. Thompson RW, Dixon AG. 1987. Population-balance models for solution-facilitated phase transformations. Proc R Soc London Ser A 413(1845):369 381. de Smidt JH, Fokkens JG, Grijseels H, Crommelin DJA. 1986. Dissolution of theophylline monohydrate and anhydrous theophylline in buffer solutions. J Pharm Sci 75(5):497501.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005

IN-LINE MONITORING OF HYDRATE FORMATION

219

19. Grijseels H, van Bloois L, Crommelin DJA, de Blaey CJ. 1983. Dissolution at porous interfaces.2. A study of pore effects through rotating-disk experiments. Int J Pharm 14(23):299311. 20. Westerberg M, Nystrom C. 1993. Physicochemical aspects of drug release.17. The effect of drug surface-area coverage to carrier materials on drug dissolution from ordered mixtures. Int J Pharm 90(1):117.

21. Liu J, Stewart PJ. 1998. Deaggregation during the dissolution of benzodiazepines in interactive mixtures. J Pharm Sci 87(12):16321638. 22. Tavare NS. 1995. Industrial crystallization process simulation analysis and design. New York: Plenum Press. 23. Ebisuzaki Y, Boyle PD, Smith JA. 1997. Methylxanthines.1. Anhydrous theophylline. Acta Crystallogr Sect C 53:777779.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 94, NO. 1, JANUARY 2005