(481b) Application of Raman and ATR-UV/Vis Spectroscopy for a Model-Free and Model-Based Active Polymorphic Feedback Control of Crystallization Processes

Introduction

The capacity of a compound to exist in more than one crystalline structure is called polymorphism. Polymorphs of the same substance can have different physical and chemical properties such as solubility, melting point or bioavailability. For this reason, determining and monitoring polymorphic transformations has become very important, especially in the pharmaceutical industries.

Many off-line techniques are available for the detection and monitoring of polymorphic transformations, such as X-Ray diffraction, differential scanning calorimetry or infrared spectroscopy. More recently, with the development of process analytical technology (PAT) tools, in situ techniques are being used routinely. The most common PAT tools used for monitoring polymorphic transformations are near-infrared (NIR) and Raman spectroscopy. However, both the techniques are rarely used for feedback control of polymorphic processes.  One of the first concepts that aimed to use information from the Raman spectroscopy to detect the formation of unwanted polymorph and change operating condition of a crystallizer was proposed by Pataki et al.‎(1) In their work Raman was used to detect the presence of the metastable form of carvedilol. In the case of the presence of the undesired form, the solution was heated up until complete dissolution and then cooled down with a different cooling rate to favour the stable polymorph. The first example of actual feedback polymorphic control on glycine was achieved by Doki et al. (2) In this approach the desired form is seeded in the correct amount during a cooling crystallization and focused beam reflectance measurement (FBRM) in combination with ATR-FTIR are used to check the total counts and the supersaturation in order to reach the desired size of the crystals and eliminate the fines. In another study the metastable form of L-glutamic acid was grown during a cooling crystallization performing supersaturation control with ATR-FTIR (3-4). A similar strategy was used by Kee to grow the metastable form of L-phenilalanine.‎(5)

All control strategies proposed in the literature so far either use the Raman system only to detect the formation of the unwanted polymorph as a trigger to restart the crystallization with a different cooling rate, or use only supersaturation control in conjunction with the suitable seed to drive the system in the phase diagram to obtain the desired polymorphic form. In this work we propose for the first time the active polymorphic feedback control (APFC), which is a combined control strategy using directly the Raman signal and the supersaturation feedback control approach for increased robustness. The Raman signal is used in the feedback control strategy to detect the presence of the polymorph contaminant and the APFC approach automatically determines the dissolution cycle needed for its elimination both in the case of seeded and unseeded systems. After the polymorph purity correction step based on the Raman signal, supersaturation control is applied to maintain the operating curve in the phase diagram between the solubility curves of the stable and metastable polymorphs hence avoiding any further contamination with metastable form. In the first part of this work a model free APFC approach is experimentally evaluated in this work using ortho-aminobenzoic acid (OABA) as the model system. In the second part the kinetic parameters of the growth and polymorphic transformation of OABA were determined through properly designed experiments in order to establish a model-based APFC that allows the elimination of the metastable form and, at the same time, maximize the size of the stable polymorph at the end of the batch.

Materials and Methodology

The model compound used for the experiments is ortho-aminobenzoic acid (or anthranilic acid), which has three known different polymorphic forms: I, II and III. (6-8) In this work only the stable form I and the metastable form II were seeded or nucleated.

Model free APFC: Raman and ATR-UV/Vis spectroscopy were used in situ for this set of experiments.  Off line analysis (XRD, Raman microscopy, DSC) were also conducted to check the purity of the material at the end of the batches. The Raman signal was related to the solid composition though a calibration-free approach while, for ATR-UV/Vis, a calibration was developed in order to measure the solute concentration from the spectra. Seeded and unseeded experiments were conducted. Different heating rates were also tested for the dissolution cycle.

Kinetic parameter estimation: Experiments were planned carefully and systematically: the different phenomena were isolated to minimize the number of parameters to estimate from each set of experiments. Growth and dissolution for both forms were estimated through seeded desupersaturation experiments either at constant temperature or with linear slow cooling. The secondary nucleation of the stable form was estimated through isothermal transformation experiments and using the dissolution and growth kinetics already estimated. Finally, primary nucleation of form I after seeding was evaluated through desupersaturation experiments with low seeds loading at high supersaturation. The initial crystal size distribution was measured using a Malvern Mastersizer.

Model based APFC: Population balance equations were solved using the method of moments. An additional condition was added in order to avoid the presence of negative moments: the dissolution terms were considered zero if the first moment became zero or negative. In this way the dissolution of each form can stop when all the material is consumed but the solution is not saturated in respect of that form. Matlab 2013 was used to solve the equations. The function ode15s was used to solve the system and a combination of fminsearch and fmincom was used to estimate the parameters. The kinetic parameters estimated were used to determine the optimal temperature profile that maximizes the size of form I at the end of the batch and allows complete dissolution of form II. Both the genetic algorithm application in Matlab 2013 and the function fminconwere used to find the optimal temperature profile that satisfies the constraints imposed.  

Results and Conclusions

The APFC approach based on the combined Raman and ATR-UV/Vis control strategy was successfully used to produce pure form I of OABA in batch crystallization experiments. For the detailed investigation of the robustness of the APFC approach different experiments were performed using both seeded of unseeded crystallizations, and varying the amount of form I and II at the moment of seeding/nucleation as well as the heating rate used for the controlled dissolution cycle. The control strategy consisted in a combination of calibration-free Raman technique and calibration-based supersaturation control using ATR-UV/Vis for concentration measurement. The Raman signal was used to identify the presence of the undesired polymorphic form and to trigger its complete dissolution. This was followed by a supersaturation control phase, based on a calibrated ATR-UV/Vis signal that maintained the operating curve in the phase diagram where the formation of the undesired polymorph is not possible. The APFC adaptively determined the dissolution cycle required for the elimination of the unwanted polymorph. It was found that the more form II is present at the moment of nucleation the higher is the temperature that will be reached in order to dissolve it all. A slow cooling rate reduced the final heating temperature and the dissolution of the stable form. The duration of the supersaturation controlled phase also strongly depended on the final temperature after that phase. The proposed APFC approach is a highly robust adaptive control strategy that can eliminate batch-to-batch variations in the polymorphic purity of crystalline products that may be caused due to variations in the nucleation composition or from varying polymorphic impurity in the seed.

In the second part of the work experiments were conducted to estimate the kinetics parameters necessary to perform a model based APFC optimization. All the parameters estimated with the proposed systematic approach present a narrow interval of 95% confidence apart from the primary nucleation of form I. That is due both to the difficulty in estimating the kinetics parameters for a stochastic process like nucleation and the small amount of experiments performed.

The results of optimization demonstrated that an initial heating step after seeding is not only required to eliminate form II but also allow a larger crystal size of form I at the end of the batch: imposing only cooling in the optimization resulted in lower crystal size, although all the metastable form naturally converted to the stable one by the end of the batch. The value of the objective function calculated was quite low compare to experimental results but this most probably is due to the uncertainty of the parameters for nucleation of form I.

References

1. Pataki, H., Csontos, I., Nagy, Z.K., Vajna, B., Molnar, M., Katona, L. & Marosi, G., Org. Process Res. and Dev., 2013, 17(3), 493-499

2. Doki, N., Seki, H., Takano, K., Asatani, H., Yukota, M., Kubota, N., 2004, 4(5), 949-953

3. Hermanto, M.W., Chiu, M.S., Woo, X.Y., Braatz, R.D., AIChE J., 2007, 53(10), 2643-2650

4. Kee, N., Tan, R.B.H., Braatz, R.D., Cryst. Growth Des., 2009, 9(7), 3044-3051.

5. Kee,N.C.S., Arendt,P.D., Tan, R.B.H., Braatz,R.D., Cryst. Growth Des., 2009, 9(7), 3052-3061

6. Jiang, S., ter Horst, J.H., Jansens, P.J., Cryst. Growth Des., 2010b, 10, 2123-2128

7. Jiang, S., ter Horst, J.H., Jansens, P.J., Cryst. Growth Des., 2010a, 10, 2541-2547

8. Jiang, S., ter Horst, J.H., Jansens, P.J., Cryst. Growth Des., 2008, 8(1), 37-43