Crystal-size distribution-based dynamic process modelling, optimization, and scaling for seeded batch cooling crystallization of Active Pharmaceutical Ingredients (API)

https://doi.org/10.1016/j.cherd.2020.10.029Get rights and content

Highlights

  • A novel d value model-based dynamic optimization of batch crystallization process.

  • Experiments with different targeted final crystal size distribution designed.

  • Cooling, stirring rate & seed amount simultaneously optimized.

  • The approach experimentally validated on different reactor vessel scales.

  • Sensitivity analysis emphasized the complex interplay of various kinetic phenomena.

Abstract

Crystallization of active pharmaceutical ingredients (API) is one of the most important and complex multiphase engineering operations in pharmaceutical manufacturing industry. The desired physicochemical properties of solid crystalline product, such as crystal size distribution (CSD), are achieved by optimizing appropriate analyzed process operating conditions. In this application-derived study, an efficient straightforward mathematical modelling approach for the d value targeted CSD (characterized by 10th, 50th, and 90th centiles d10, d50, and d90) optimization of the API fesoterodine fumarate (FF) batches in different solvent mixtures on reactor dimension scales from 0.1 to 15 L is presented. The model is based on energy, mass, and population balance equations, the thermodynamic system equilibrium between solute/solution, and the kinetics of nucleation, crystal growth, and crystal agglomeration. In the first set of two experiments, the ability of the model to predict final CSD under chosen operating conditions was validated applying particular previously estimated kinetic parameters. An excellent statistical agreement between predicted and experimental CSD results was observed. Furthermore, the utility of the model to determine suitable operating conditions for the formation of FF crystals with d value defined CSD is presented. Two additional experiments were designed where stirring, cooling rate, and the amount of seed were optimally regressed. Good agreement between targeted and experimental CSD was shown and depending on the chosen vessel unit, mixing & cooling rates had the strongest relative impact on CSD. Algorithm may be beneficially utilized early during API industrial technological development, intensification, and scale-up, or transferred to continuous flow.

Introduction

Crystallization of active pharmaceutical ingredients (API) is one of the most important separation and purification operations in pharmaceutical industry. Desired physicochemical properties of solid crystalline product, such as crystal size distribution (CSD), crystal morphology, polymorphic form, and purity, which all have a tremendous impact on further processing and biopharmaceutical properties of the medicinal products, are manipulated by API crystallization. In order to continuously supply high quality API with defined properties, the operating conditions of crystallization processes are therefore subject to rigorous optimization and strict control during manufacturing. This however requires a detailed mechanistic understanding of the process (Myerson, 2001; Tung et al., 2009; Tung, 2013; Shekunov and York, 2000).

The industrial crystallization process development has been acknowledged in the literature as one of the most challenging with respect to optimization and scale-up, as it depends on the complex interplay of several thermodynamic and scale-dependent kinetic, transport, and hydrodynamic phenomena in a heterogeneous system (Kougoulos et al., 2006; Chen et al., 2011). A thorough characterization of the solubility equilibria for various polymorphic and pseudopolymorphic crystal forms constitutes the basis of crystallization process development. However, the crystal properties are pronouncedly dependent on initial crystal population (seeds) and various kinetic mechanisms, which are heavily influenced by supersaturation and mixing. Especially the latter is highly affected by the process scale, as mixing time, mixing intensity, and mixing distribution may change considerably upon scale-up (Tung, 2013). Very often a trade-off is required upon scale-up to neglect some of the product properties in favour of others which are more critical for final product quality (Beckmann, 2013).

Mathematical modelling is a useful tool for design, optimization, and control of crystallization processes. However, due to the complexity of the process, it has not yet been generalized to a similar extent as modelling of other unit operations. Its potential is thus much greater than its current contribution (Myerson, 2001; Tung et al., 2009). Model-based dynamic optimization methodology of crystallization processes has already been thoroughly evaluated in the literature. Therein, a validated mathematical model is utilized to estimate suitable values of certain operating conditions (for example temperature and/or antisolvent addition profile) of the process that lead to formation of crystalline product with targeted properties. These are usually targeted indirectly in order to produce crystals with high filterability more suitable for further processing, for example to achieve maximal final mean crystal size (Lang et al., 1999; Choong and Smith, 2004; Yang et al., 2007; Nowee et al., 2008; Nagy et al., 2008; Su et al., 2015; Xie and Schenkendorf, 2019) or minimize the extent of nucleation (Rawlings et al., 1993; Matthews and Rawlings, 1998; Hsu and Ward, 2013). Multi-objective optimization targeting several properties concomitantly, often including also minimization of final CSD coefficient of variation and minimization of process time, has been presented in several studies with successful experimental validation (Chung et al., 1999; Togkalidou et al., 2004; Hemalatha et al., 2018; Pal et al., 2019; Patience et al., 2004; Sarkar et al., 2006, 2007; Nowee et al., 2007; Trifkovic et al., 2008; Lindenberg et al., 2009; Ridder et al., 2014; Acevedo et al., 2015).

Examples where final properties are targeted directly, which is most often required in medicinal compounds’ manufacturing due to very tight product specifications, are somewhat fewer. Usually, desired mean size (Patience et al., 2004) or the whole final CSD are targeted (Worlitschek and Mazzotti, 2004; Aamir et al., 2010; Nagy et al., 2011; Majumder and Nagy, 2013; Ridder et al., 2016; Szilagyi and Nagy, 2019). Specifications for solid API however very often utilize d10, d50, and/or d90 values, which are the crystal sizes that correspond to the tenth, fiftieth and ninetieth centile of the cumulative CSD, respectively (Zhigang et al., 2010). While it is likely that similar results would be obtained in the mentioned studies had they applied d values instead of mean sizes/whole CSD, no such studies have been presented to the best of the authors’ knowledge. Additionally, different types of crystallization processes (cooling crystallization, antisolvent crystallization, precipitation) in different types of reactors (batch, semi-batch, continuous) have been extensively studied with respect to model-based dynamic optimization, but none of these studies validated the presented methodology on several different reactor scales, despite the fact that development, optimization, and scale-up of industrial crystallizations require a thorough process understanding on several reactor scales.

In this paper, we present our model-based scale-independent optimization approach for targeting final CSD. Batch crystallization of polymorphic form I of fesoterodine fumarate (FF; Fig. 1a and b), an API used for treatment of overactive bladder, was chosen as the model system. Thermodynamic, transport, and kinetic phenomena of FF crystallization from different mixtures of methyl ethyl ketone (MEK) and cyclohexane (CHX) have been examined in our previous studies and the mathematical model was validated on a single reactor scale (Trampuž et al., 2019, 2020). The work presented in the paper is graphically outlined in Fig. 2. In the first part of the paper, the cooling crystallization model using previously estimated values of kinetic parameters is experimentally validated by performing two experiments on different reactor scales with predefined operating conditions. Experimental and simulated final CSD are compared to validate the transferability (extrapolation) of the model to different reactors. The results are supported by thorough model-based analysis of physical transport and hydrodynamic phenomena of the system which are subject to change on different scales, and must be considered for a successful industrial scale-up campaign (Kougoulos et al., 2006; Schmidt et al., 2004). In the second part, the d value model-based optimization (MBO) procedure is presented and applied to estimate suitable operating conditions to achieve desired final CSD. The optimized conditions are experimentally validated on two different reactor scales as well through comparison between experimental and desired CSD. In the final part of the paper, a comprehensive sensitivity analysis of the FF batch cooling crystallization process is presented to understand the relationships between various operating conditions and final product d values and test the robustness of the model.

Section snippets

Materials

FF in form I (> 99.0 area % UPLC purity), MEK, and CHX were obtained from Lek d.d. Experiments were performed in three different reactors: 100-mililiter glass batch reactor EasyMax 102 Advanced Synthesis Workstation with pitched blade turbine impeller (four blades under 45° angle) and aluminium thermostat (Mettler Toledo) (reactor A), 2-liter glass batch reactor AP01-2 with pitched blade turbine impeller (four blades under 45° angle), integrated into a RC1e reactor system (Mettler Toledo,

Validation experiments

Fig. 6 and Table 4 show the comparison of experimental and simulated final CSD for Exps. 1 and 2. Crystals of similar size were obtained in both experiments and very close agreement between model simulations and experiments is observed. Taking into account sum of least squares method for quantitative assessment of the goodness of fit, better agreement was observed in Exp. 1. Experimental and simulated d50 values were however closer in Exp. 2.

In order to understand the impact of different

Conclusion

To summarize, the present study describes a simple and efficient approach to simultaneously optimize the values of some of the most important batch crystallization process operating conditions in order to produce final crystals with desired CSD. The latter is simply expressed as commonly used d values, more specifically their discretized counterparts (abbreviated as dc values). By discretizing the length coordinate fine enough, the differences between d and dc values may be negligible. The

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The authors declare no competing financial interest.

Declaration of Competing Interest

The authors report no declarations of interest.

Acknowledgements

The authors gratefully acknowledge Dr. Vesna Stergar, Prof. Dr. Zdenko Časar and Mr. Pavel Drnovšek from Lek d.d. for their support of this work. The financial support for conducting the research and preparation of the article was provided by Slovenian Research Agency (grant program number P2-0152) and Lek d.d.

References (54)

  • D. Sarkar et al.

    Multi-objective optimization of seeded batch crystallization processes

    Chem. Eng. Sci.

    (2006)
  • B.Y. Shekunov et al.

    Crystallization processes in pharmaceutical technology and drug delivery design

    J. Cryst. Growth.

    (2000)
  • D. Shi et al.

    Predictive control of particle size distribution in particulate processes

    Chem. Eng. Sci.

    (2006)
  • Q. Su et al.

    Mathematical Modeling, Design, and Optimization of a Multisegment Multiaddition Plug-Flow Crystallizer for Antisolvent Crystallizations

    Org. Process Res. Dev.

    (2015)
  • B. Szilagyi et al.

    Model-based analysis and quality-by-design framework for high aspect ratio crystals in crystallizer-wet mill systems using GPU acceleration enabled optimization

    Comput. Chem. Eng.

    (2019)
  • M. Trampuž et al.

    Crystallization of fesoterodine fumarate active pharmaceutical ingredient: Modelling of thermodynamic equilibrium, nucleation, growth, agglomeration and dissolution kinetics and temperature cycling

    Chem. Eng. Sci.

    (2019)
  • M. Trampuž et al.

    Process Analytical Technology-based (PAT) Model Simulations of a Combined Cooling, Seeded and Antisolvent Crystallization of an Active Pharmaceutical Ingredient (API)

    Powder Technol.

    (2020)
  • X. Xie et al.

    Stochastic back-off-based robust process design for continuous crystallization of ibuprofen

    Comput. Chem. Eng.

    (2019)
  • D. Acevedo et al.

    Multiobjective Optimization of an Unseeded Batch Cooling Crystallizer for Shape and Size Manipulation

    Ind. Eng. Chem. Res.

    (2015)
  • W. Beckmann

    Crystallization: Basic Concepts and Industrial Applications

    (2013)
  • J. Chen et al.

    Pharmaceutical Crystallization

    Cryst. Growth Des.

    (2011)
  • S.H. Chung et al.

    Optimal Seeding in Batch Crystallization

    Can. J. Chem. Eng.

    (1999)
  • EasyMax - synthesis made simple, Manuf. Chem. (n.d.)....
  • A. Garg et al.

    Subspace Identification-Based Modeling and Control of Batch Particulate Processes

    Ind. Eng. Chem. Res.

    (2017)
  • C. Hsu et al.

    The Best Objective Function for Seeded Batch Crystallization

    AIChE J.

    (2013)
  • E. Kougoulos et al.

    Process modelling tools for continuous and batch organic crystallization processes including application to scale-up

    Org. Process Res. Dev.

    (2006)
  • Y. Lang et al.

    Dynamic Optimization of a Batch Cooling Crystallization Process

    Ind. Eng. Chem. Res.

    (1999)
  • Cited by (20)

    • Evaluation of the kinetics of unseeded batch cooling crystallization using population balance modeling: Sucrose and KNO<inf>3</inf> case studies

      2023, Journal of Industrial and Engineering Chemistry
      Citation Excerpt :

      Thus, methodologies to obtain crystallization kinetics for batch crystallization experiments have been increasingly studied. [1–4,9–13]. Kinetic parameters of both nucleation and growth have been evaluated by fitting conservation equations to measured variables, such as concentration and CSD, during crystallization [2,3,10–13], through optimization method such as least square, generic algorithm, dynamic optimizations, and so on [14–18]. Since Larson and Randolph [19] introduced, most widely used conservation equations for evaluating nucleation and growth of crystals is the population balance equation (PBE) [2–3,10–13,20–24].

    • Measurement and correlation of 4,4′-dihydroxydiphenylsulfone solubility in ten neat organic solvents at temperatures from 278.15 to 313.15 K

      2022, Journal of Chemical Thermodynamics
      Citation Excerpt :

      It is well known that solvent crystallization is a fundamental operating unit in the purification process of solid products. The crystallization process determines the quality of the final crystal product, such as crystal habit, crystal shape, size distribution, purity, etc [4–9]. Solubility is vital for optimizing the crystallization process as a substance's most basic thermodynamic property.

    View all citing articles on Scopus
    View full text