Industrial Perspectives of Pharmaceutical Crystallization - American

Nov 27, 2012 - ABSTRACT: Industrial practice of pharmaceutical crystallization is reviewed in ... tinuing supply of active pharmaceutical ingredients ...
0 downloads 0 Views 3MB Size
Review pubs.acs.org/OPRD

Industrial Perspectives of Pharmaceutical Crystallization Hsien-Hsin Tung* Process Engineering, Global Pharmaceutical R&D, Abbott Laboratories, North Chicago, Illinois 60064, United States ABSTRACT: Industrial practice of pharmaceutical crystallization is reviewed in this manuscript through three levels of perspectives. The first level addresses the thermodynamic properties, which include basic properties such as solubility, crystal form, and morphology. The second level addresses the transport properties and rate behavior of crystallization, specifically kinetics of crystallization and mixing within a uniform suspension environment. Factors such as seeding, control of supersaturation, mixing time, and mixing intensity are discussed at this level. The third level addresses the overall process dynamics and scale-up issues. The main focus is on the rate and transport phenomena in a nonuniform suspension environment. Topics of mixing distribution and crystallizer design are addressed here. Specific examples are cited to demonstrate the impact of these factors on the design and performance of crystallization process.

1. INTRODUCTION Crystallization has been the most important separation and purification process in the pharmaceutical industry throughout its history1,2 and has generated significant research and development activities in past decades.3−6 In the pharmaceutical industry, the issue of better control, desirable in and of itself, is reinforced by the need to satisfy the regulatory authorities so that a continuing supply of active pharmaceutical ingredients (APIs) of high and reproducible quality and bioavailability can be delivered for formulation and finally to the patient. The “product image” (properties, purity, etc.) of this medicine must be the same as that used in the clinical testing carried out to prove the product’s place in the therapeutic marketplace. These requirements demand efficiency of the crystallization process to meet time and supply constraints, as well as scalability from laboratory to factory and batch-to-batch consistency and robustness throughout all phases of drug development. In this review, an industrial approach to meet these constraints and requirements is presented. The approach is divided into three levels of perspective with increasing degree of complexity. The impact of attributes in each level on final crystal qualities is discussed. Selected cases based upon the author’s experience are cited for the purpose of demonstration.

where a is the activity of compound of interests in solution, which is directly related to the amount of compound dissolved, solubility, and γSAT activity coefficient, ΔfusS is the i.e. xSAT i i entropy of fusion, R is the Boltzmann constant, T is the temperature, and Tm is the melting point of the compound. In this equation, it is assumed that the difference between heat capacities of the compound as liquid and as solid is negligible. The purpose of presenting this equation is to summarize factors which can affect the solubility. As shown in this equation, temperature and entropy of fusion, which is equivalent to the difference of chemical potentials between the solid and the liquid of a particular compound, can directly affect the solubility. In addition, the solvent composition of mixed solvents, as well as other minor components or impurities in the solution, can affect the activity coefficient. Many interesting phenomena have been observed. It is common to observe that solubility is maximized in the middle range of solvent composition of a solvent/antisolvent mixture. Also, the solubility of a hygroscopic compound in an alcoholic solvent can be very sensitive to the water level (at few hundreds of ppm) in the solvent. Chemical structure and salt forms (or cocrystals) of the compound can affect entropy of melting and activity coefficient and, hence, the solubility. Solubility measurement and its prediction have received a lot of attention in recent years. Predictive and correlative models, for example the NRTL-SAC model7 which is considered to be the benchmark currently, have been developed and applied to solve practical industrial applications. As reported in the cited study,8 the NRTL-SAC model is able to predict solubilities of four different compounds in multiple solvents with a wide range of hydrophobicity and hydrophilicity, and with solubility values ranging over six-orders of magnitude of reasonable accuracy. Solubility models, such as the NRTL-SAC method, are now routinely used in the pharmaceutical industry in recent years. Extension to a wider basis, for example drug solubility in

2. LEVEL I: BASIC THERMODYNAMIC PROPERTIES At this level, the basic thermodynamic properties will be addressed. Attention will be given to solubility, polymorphs, including solvates, and morphology. 2.1. Solubility and Solubility Map. Understanding the solubility behavior is the first step toward the successful development of crystallization processes. For solution crystallization, the solubility of a chemical compound is simply the equilibrium (maximum) amount of this compound that can dissolve in a specific solvent(s). Hence, the solid solute and dissolved solute are at equilibrium under this condition. Mathematically, the solubility can be expressed as shown in eq 1. ln a = ln xiSAT·γiSAT

Δ S⎛ T ⎞ = fus ⎜1 − m ⎟ ⎝ R T ⎠ © XXXX American Chemical Society

Special Issue: Polymorphism and Crystallization 2013 Received: August 26, 2012

(1) A

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

to a lower solution concentration (or supersaturation) region which favors the formation of crystalline solid. At present, there are no theoretical models available which can quantitatively estimate the solution concentration corresponding to the type of secondary phase with industrial significance, based upon author’s knowledge. The solution concentration (or supersaturation) can also affect particle size. As another example, Figure 2 shows the images of particles formed under two different operating modes of an antisolvent crystallizationforward addition and reverse addition. In the forward addition mode, where the antisolvent is added to the solvent containing dissolved batch, larger particles are generated because supersaturation is low. In the reverse addition mode, where the batch solution is added to the antisolvent, smaller crystals are generated because supersaturation is high. 2.2. Crystal Forms (Polymorphs and Solvates). Polymorphism is a common phenomenon observed in pharmaceutical crystallization. Polymorphic crystals are the same molecular species, but with different crystal structures. The impact of polymorphism is beyond the level of crystallization because it could affect the downstream formulation and drug bioavailability. The search for possible polymorphs or solvates, which could be theoretical, empirical, or a combination of both, is a subject undergoing extensive research and development efforts3−6,9 It is not unheard of that a new crystal form of a drug candidate appears in the late phase of drug development or after the drug is approved and distributed on the market.10 Currently in the industry, the search for polymorphs or solvates is typically empirical. No particular empirical methods, such as cooling, antisolvent, evaporation, or salt formation, have been demonstrated to be superior to other empirical methods. The focus of the search could vary at different phases of drug development. In the early phase, the focus would be to find “one” stable form suitable for animal and human clinical evaluation. In the later phases, the focus would be to find the most stable crystal form for commercial development. There are, of course, exceptions to this statement. If the crystal form used in the early phase of drug development still meets the commercial API manufacturing and formulation criteria, it may still be used in the later phase of drug development. As mentioned above, polymorphic crystals have different crystal structures. They have different X-ray diffraction patterns and Raman spectra. In addition, they should have different physical attributes, for example morphology, solid density, heat capacity, heat of melting, melting point or decomposition temperature, solubility, etc.11 One interesting development recently is the identification of the specific crystal form via the three dimensional (3D)-crystal morphology of a crystal.12 In reference 12 the 3D-crystal morphology is constructed through multiple 2D-crystal images at different focal depths. After constructing the 3D-crystal morphology of the crystal, the angular patterns among all crystal surfaces on the crystal are calculated and compared to the angular patterns of all known crystal forms of the compound. Since each crystal form has its own unique angular patterns, this method can identify the specific crystal form on the simple basis of the crystal morphology. The correlation between crystal form and morphology using the microscope could offer potential advantages over other established techniques such as Raman or powder X-ray diffraction methods, such as simplicity of measurement, a significant reduction of equipment and maintenance cost, and a better sensitivity

polymers/surfactants for lipid-based or solid dispersion formulation, could be of great value for the future. For the design of a crystallization process, at the beginning the conditions are chosen so that all API or intermediate solid can be dissolved. At the end, the conditions are chosen so that the majority of the batch is crystallized. It is desirable that impurities such as reaction byproducts or other undesired species remain soluble in the mother liquor at this point. Generally, much less effort is spent in the industry to quantify the solubility of impurity in solvents. One key reason is the amount of impurities isolated being insufficient for the solubility measurement. The other reason is the diversity of impurities within one compound. Instead, simple rules based upon other information are applied in guiding the selection of solvents, for example the retention times of impurities from the liquid chromatogram or the chemical structures of impurities if available. Polar solvents such as water or alcohol are chosen for the rejection of early-eluting or polar impurities. Nonpolar solvents such as ester or alkane are chosen for the rejection of late-eluting or nonpolar impurities. It should be noted that the solubility difference between the desired compound and impurities may not be the only factor in affecting the rejection of impurities. Other factors such as solid solution, inclusion/occlusion during the crystallization, or absorption of impurity on the crystal surface may affect the rejection efficiency as well. More details can be found in reference 1. Beyond a specific knowledge of solubility, it is also beneficial to construct an empirical qualitative solubility map as shown in Figure 1. This map is drawn mainly on the basis of the author’s

Figure 1. Solubility map showing the relationship between solution concentration and physical characteristics of the secondary phase.

empirical observation of pharmaceutical compounds in the past. It correlates the degree of solution concentration, i.e. supersaturation, to the type of secondary phase, i.e. oil, amorphous solid, or crystalline solid which can be encountered during the crystallization, thus providing a practical guide for the development of crystallization. For example, if the objective is to generate fine nuclei of a particular API and the observed secondary phase under the experimental conditions produces predominantly large crystals, the experimental conditions need to be adjusted to operate at a higher solution concentration (or supersaturation) region which creates a higher degree of nucleation. On the other hand, if the observed secondary phase is oil or amorphous solid, the experimental conditions need to be adjusted B

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

Figure 2. (A) Forward addition (left) and (B) reverse addition (right) of an antisolvent crystallization.

of solubility is needed to identify the feasible paths for crystallization. The above discussion of polymorph conversion can be extended to the case of solvates. Specifically, above certain solvent composition, one solvate form is more stable (form I solvate). Below that solvent composition, another solvate form is more stable (form III solvate). Form II may be an anhydrate throughout the entire solvent composition range. Similarly, a comparable diagram can be established to describe the conversion of crystal forms during drying or humidification. Above a certain level of humidity, one hydrate is more stable. Below a certain level, another hydrate or anhydrate is more stable. Crystal packing calculation13 can also be used to assess the crystal structures and facilitate the interpretation of form conversion during slurrying or drying from the molecular levels. 2.3. Morphology: Crystal Habit. Morphology or crystal habit is another important and interesting physical attribute because of its potential impact on downstream filtration, drying, and formulation. Contrary to crystal form, crystal habit typically refers to the appearance of crystals. It does not reflect the internal structure of crystals. We will be using three qualitative terms below to describe the shape of crystals. The first type is called needle-like because it has only one key dimensional length. The second type is called plate-like because it has two key dimensional lengths. The third type is called rod-like or cube-like because it has three key dimensional lengths. From the shape of a crystal, it is possible to infer the surface area and relative growth rates of different crystal surfaces. For needle-like crystals, the surface for crystal growth is primarily on the two tips. The surface on the needle has a much slower growth rate. For plate-like crystals, the surface for crystal growth is at the edges. The surface on the plate has a much slower growth rate. For rod-like or cube-like crystals, all crystal surfaces grow at comparable rates. Inferring the relative growth rates on different crystal surfaces has practical impact on the crystallization process development. In general, for needle-like crystals, more seed is needed since it has less surface area for crystal growth, despite the overall surface area per unit mass of crystals being high. On the other hand, for rod-like crystals, less seed is needed even though the overall surface areas per unit mass of crystals is low. Needle-like or plate-like crystals could create operational complications such as poor cake filtration and washing, breakage of particles to produce fines during handling, and poor solid flow for formulation. In practice, it is always desirable to grow “thicker” crystals. Various factors can affect the crystal morphology.

in detecting crystals of undesired forms within the matrix of crystals of desired form. Solubility, which is essentially chemical activity as shown in eq 1, is also a very useful and practical measure associated with polymorphs. To exemplify this point, as shown in Figure 3,

Figure 3. Relationship of solubility curves of polymorphs.

form I has a lower solubility than that of form II throughout the entire solubility region. Forms I and II are called monotrophic polymorphs, i.e. form I is more stable than form II throughout the entire solubility region. Figure 3 also shows that form I has a solubility higher than that of form III at the left region of the figure, but has a solubility lower than that of form III at the right region of the figure. Forms I and III are called enantiotropic polymorphs, i.e. polymorph I is more stable at the higher solubility region and is less stable at the lower solubility region. Given the knowledge of solubility relationship between different crystal forms, it is easy to conceive that, for crystallization path A, form II could be observed initially, then form II would convert to form I and eventually could turn into form III. The sequence of crystal forms observed will be form II → form I → form III. By applying the same rationale for crystallization path B, the sequence of crystal form could become form III → form II → form I. The reverse logic can also be applied. By knowing the conversion pattern of different crystal forms, it is feasible to reconstruct the solubility relationship. In either direction, the knowledge C

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

release of supersaturation, and minimizes the nucleation. It can reduce inclusion or occlusion during crystallization and maximize the rejection of impurities. There are numerous industrial case studies to support all these claims.1,16

These factors can be crystal structure, solvents, additives, impurities, and supersaturation or desupersaturation rates during crystal growth and dissolution periods, etc. Additionally, it is shown that multiple heat/cool cycles coupling with wet milling1 at each cool cycle or without wet milling14 can be employed to modify the crystal morphology as well. This approach has been practiced over the years in the industry in modifying the crystal morphology, without modifying the solvents or crystal forms or charging additives during the crystallization due to chemical purification or bioavailability requirements. At present, industry has built a significant amount of know-how of this approach, with a limited understanding of fundamental mechanism. Further research can help to fill the gap in fundamental mechanism.

Figure 4. Methods of seed generation and corresponding size ranges.

3. LEVEL II: CRYSTALLIZATION KINETICS AND MIXING At this level, the focus will be on the transport properties and rate behavior of crystallization. In addition, the entire system is assumed to be in well-mixed state, i.e. homogeneous suspension. The kinetics of crystallization can be described by the population balance equation below:15 (nVG) ∂(nV ) + = ∂t ∂L

Generation of seed can be classified into two approaches: wet and dry as highlighted in Figure 4. Traditionally, seed is generated by the dry milling approach. For the dry approach, the current batch of solid API (or intermediate) is dry-milled. The bulk of the batch is used for downstream formulation and portion of the dry-milled API (or intermediates) is retained as seed for the next batch. Depending upon the API particle size specification, pin mill and jet mill are two common types of dry mills used as shown in Figure 4. Dry milling is convenient in the early phase of drug development since dry milling operations consume less resource for process development. However, there are many disadvantages associated with dry milling. For example, material can melt or form a compacted layer within the mill chamber. Solvates or hydrates can be desolvated or dehydrated. Crystals can lose their crystallinity and become amorphous. The milled particles can interact with the mill metal surface. Additionally, dry milling is a significant contributor to the overall cost of manufacturing of APIs. Given these considerations, wet seed is a preferred method for the later phases of drug development. The advantages of wet seed for crystallization include controlled seed crystal form, tunable seed size and amount, and robustness of scale-up. Wet seed can be generated by wet milling the API in a saturated slurry. Depending upon the seed size requirement, multiple choices of wet milling technologies17 can be employed, for example media mill for submicrometer particles, ultrasound device or rotor/stator homogenizer for micrometer particles. Since particle size is reduced during the wet milling step (as in the dry milling), this is a top-down approach. Another approach for generating the wet seed is called the in situ seed generation approach.18,19 In this approach, API is dissolved in a suitable solvent mixture. Fresh solid nuclei, which are named in situ seed, are formed under a supersaturated condition through a controlled mixing environment. Rapid precipitation through a mixing device such as an impinging jet mixer20,21 to generate fine particles also falls into this category. In situ wet seed generation clearly obviates the analytical testing needs for longterm chemical and physical stability of dry seed material. Additionally, it is very flexible over other seed generation methods in terms of seed amount since seed is generated on demand. One caution of the in situ seed generation approach is the control of the crystal form because metastable forms, amorphous solids, or oil droplets can form initially. If the stable crystal form is desired for the seed, an additional aging with intensive mixing could be applied to facilitate the conversion of metastable crystal forms to the stable crystal form.22 Adding a pinch of dry/wet seed of the stable crystal form prior to the generation of in situ seed could be a useful trick as well. In comparison to

∑ Q innin − ∑ Q outnout

initital conditions:

n(L , t = 0), C(t = 0) rate equations: G=

dL = kg ·S g dt

B = k b·N x ·μ2yor 3 ·S b at L = 0 S=

C Csolubility

or C − Csolubility (2)

where n represents particle size distribution which is a function of t time and L particle size/length; V is the volume of batch; Qin and Qout are respectively in-flows and out-flows to the crystallizer; G and B are crystal growth and nucleation rate expressions which are functions of kg crystal growth rate constant, kb nucleation rate constant, g crystal growth exponential factor, b nucleation exponential factor, N agitator speed, μ2 second moment of particles, μ3 third moments of particles, x, y are exponential factors, C solution concentration, Csolubility solubility, and S supersaturation. As described in this equation, the key contributors influencing the particle size distribution are initial particles, i.e. seed, and crystal growth and nucleation rates which are functions of supersaturation and mixing. 3.1. Seed. The criticalness of seed for crystallization cannot be overemphasized. Seed impacts all aspects of crystallization, including crystal form, particle size distribution, crystal growth and nucleation rates, and final product purity. Seeding the batch with the desired crystal form can mitigate the risk of forming the undesired crystal form. Even if the desired crystal form is metastable, seeding the batch with the metastable form could still be a feasible approach. For a growth-dominant crystallization process where nucleation is minimized, the final particle size can be calculated straightforward based upon the seed amount, seed size and the amount of solid grown onto the seed crystals. Also, with a sufficient amount of seed, it accelerates the D

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

To gain good control of supersaturation, some quantitative measures of crystal growth and nucleation rate constants are needed. Due to the diversified nature of API (or intermediate) crystals, nucleation and crystal growth rates can vary drastically over several orders of magnitude. At the current time, it is not possible to quantitatively predict the crystallization behavior on the basis of theoretical models. To address this limitation, a model-based experimental design (MBED) methodology for crystallization23 has been developed. The basic procedure is outlined as follows: (1) Estimate crystallization and nucleation kinetic parameters by fitting existing experimental data, including particle size distribution and solution concentration profiles, during the crystallization via population balance eq 2. (2) Design seed (amount, size, or surface area) and a crystallization path (cooling, antisolvent/reactant addition profiles, etc.) to minimize uncertainty in parameters. The choice of seed and the crystallization path may be driven toward a growth-dominant, nucleation-dominant, or a hybrid process, depending on uncertainties in the kinetic parameters. (3) Conduct additional experiments and collect measurements. Repeat steps 1, 2, and 3 sequentially until desired accuracy in parameters is achieved. (4) Optimize seed and the crystallization path to meet design targets. As demonstrated in the cited study,23 in using this methodology only three experiments were needed to fully optimize a cooling crystallization process with a high confidence level of kinetic parameters. In the cited reference,23 for example, the uncertainty of crystal growth exponential factor g is reduced from ±30% in the first experiment to ±11% in the second experiment. Similarly, the uncertainty of nucleation exponential factor b is reduced from ±50.4% in the first experiment to ±12.4% in the second experiment. In the third experiment where the cooling profile and seed are optimized, the ratio of mass grown on the nucleated particles over mass grown on the seed is reduced from 61% in the first experiment to 21% in the third experiment. This is a significant saving of resources. The limitations of MBED methodology reside mainly in the validity of the population balance equation and the accuracy of experimental particle size distribution and solution concentration data. The population balance equation, eq 2, may not be adequate to describe all the observed crystal behaviors such as particle aggregation or breakage. The measured experimental particle size data by the in-line particle size analyzer probe may not be representative due to the variation of hydrodynamic flow pattern in the vicinity of the probe during the crystallization. Despite these limitations, MBED methodology still provides more valuable quantitative information, beyond qualitative trending of particle size distribution. To the author’s knowledge, MBED methodology has been successfully implemented in practice. The exact experimental procedures and numerical algorithms for parameter estimation and optimization may vary from user to user and from company to company. Another approach to control the supersaturation is modelfree feedback control24,25 of concentration or nucleation. One advantage for this approach is that there is no modeling requirement. However, the quality of control will depend directly on the quality of the data collected during crystallization. In actual commercial implementation, this will be a nontrivial task because the real-time control will impact the real-time product quality. Additional measures to ensure the robustness of feedback control,

the top-down wet-milled seed approach, in situ seed generation is the bottom-up approach since solid nuclei are formed from a supersaturated solution via a primary nucleation mechanism initially and a secondary nucleation mechanism afterward. Since supersaturation and controlled mixing are applied to generate the in situ seed, a recirculation loop around a crystallizer coupled with an in-line mixer provides a suitable scheme for implementing this technique.19 As shown in Figure 5, under

Figure 5. Generic diagram showing a crystallizer with an external recirculation loop and in-line mixer (or wet mill).

the reverse-addition operation mode, the antisolvent is circulated through a crystallizer external loop which includes a highspeed rotor/stator homogenizer. The dissolved batch is fed into the external recirculation loop of the crystallizer. The degree of supersaturation can be controlled by the feed rate and the recirculation flow rate in the crystallizer. The mixing intensity can be controlled by the rotor/stator configuration and the rotor tip speed. In the reference 19, this approach successfully generates the final particle size at a mean size of 9−10 μm, which is much smaller than that obtained by the direct wet milling approach. Also, it eliminates the need for dry milling. The recirculation scheme can certainly be applied to forward addition mode operation where antisolvent (or reactant) is charged into the loop and the solvent containing the batch is recirculated, or simultaneous addition mode operation where multiple streams are charged into the loop simultaneously (see also section 4.2). 3.2. Supersaturation. Supersaturation affects both crystal growth and nucleation rates, which in turn impact the particle size distribution. A higher level of nucleation leads to smaller particles and vice versa. Also, a high degree of nucleation rate over crystal growth rate due to a high degree of supersaturation can lead to poorer rejection of impurities.1 Given these considerations, control of supersaturation, coupled with the utilization of proper seed, to maximize crystal growth and minimize nucleation is generally preferred. It should be pointed out that (on the basis of the author’s personal experience), if a higher degree of nucleation is created by a higher level of mixing intensity instead of supersaturation alone, it does not necessarily affect the product purity. E

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

for example prevention of probe fouling or redundant measurement, would be highly desirable. 3.3. Mixing Time. Given the complexity of mixing effects on crystallization kinetics, we will provide a simple picture first and address two key factors at this level where the entire system is assumed to be uniformly suspended. More elaboration on the scale-up impact will be addressed in section 4. The first key mixing factor is mixing time. Here, the mixing time is loosely defined as the time required to reach 95% (or any specific percentage) homogeneity of multiple streams upon mixing. These streams could be solvent and antisolvent, or drug in one solvent and reactant in another solvent, e.g. in making salts, etc. Since mixing of these streams creates supersaturation, the time of mixing to reach composition homogeneity will clearly affect the subsequent nucleation and crystal growth rates. Let us further consider two dimensionless numbers1 below: Danucleation =

Figure 6. Time for release of supersaturation as function of seed loading.

τmixing τinduction

Dacrystallization =

apparent fluid or agitator tip velocity. These terms will be used interchangeably to represent the mixing intensity throughout this manuscript. Impact of mixing intensity on crystallization kinetics is multifaceted. It can affect the mixing time, particle breakage, primary and secondary nucleation rates, and also crystal growth rate. In the following discussion, we will focus on the mixing intensity impact on particle breakage and nucleation because of its relevance to wet seed generation. A comprehensive discussion of this subject can be found in the reference 26. Figure 7 illustrates the mixing intensity impact on particle breakage in a saturated slurry27 using a hydrodynamic cavitation

τmixing τrelease of supersaturation

(3)

where τmixing is the mixing time, τinduction is the induction time for primary nucleation, and τrelease of supersaturation is the time required to release, for example, 95% or other levels of the supersaturation. Danucleation ≪ 1 means a complete mixing before the nucleation is achieved. This would be the case for impinging jet crystallization20,21 where two streams are mixed rapidly to generate a high degree of supersaturation before solid (or oil droplets) spontaneously nucleates. Similarly, Dacrystallization ≪ 1 means a complete mixing is achieved before the supersaturation is released. This would be the case for crystallization with relatively slow crystallization kinetics in releasing the supersaturation. Since the order of mixing time in a crystallizer is generally available, it is straightforward to learn if the crystallization system could be sensitive to mixing by comparing the induction time for nucleation or time for release of supersaturation. Mixing time could be affected by multiple factors,26 including mixing intensity (section 3.4), scale (section 4.1), and mixer geometry (section 4.2) etc. Similarly, induction time for nucleation and time for release of supersaturation can be affected by multiple factors1 as well. For example in Figure 6, at a seed level from 0.2% to 1%, it needs more than 30 min to release the supersaturation. Since mixing time in a stirred tank is shorter (on the order of few minutes), the local supersaturation at the antisolvent or reactant addition points is less likely to affect the overall crystallization kinetics. In other words, the entire solution can be treated as homogeneously mixed with regard to the crystallization kinetics. However, when seed level increases to 5% and 15%, the solution supersaturation is released much faster. As a matter of fact, their supersaturation release profiles overlap with each other. Since this time scale is at a time scale similar to that of mixing, at 5% and 15% seed loadings it is reasonable to conclude that the release of supersaturation is limited by the mixing, instead of crystallization kinetics. In these cases, the local mixing can affect local crystal growth and nucleation rates and thus affect the overall crystallization performance. 3.4. Mixing Intensity. The second key mixing factor is mixing intensity. Here mixing intensity is loosely defined as power (energy dissipation rate) per unit volume, shear rate, or

Figure 7. Correlation between steady-state particle size and fluid velocity.

mill. In Figure 7, different shapes of data points represent different compounds. Different colors for each shape represent data of the same compound under different mill configurations. Very interestingly, there is an excellent correlation between the apparent fluid/tip velocity (v in m/s in the X-axis) and the particle size at steady state (dss in the Y-axis). Furthermore, this correlation appears to be applicable (on the basis of the author’s experience) to other types of mills as well, i.e. rotor/stator homogenizer and stirred tank agitators. It should be noted that this curve is not aligned with the results from a media mill where F

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

trend of mixing time (order of magnitude) as a function of scale (order of magnitude). As shown in Figure 8, the mixing time will vary from fractions of seconds to minutes and up, when the scale is increased from microscale to macroscale. For example, in the laboratory where the size of crystallizer is typically up to hundreds of milliliters, the mixing scale resides in the mesoscale domain (or microscale if using special micromixing devices), and the mixing time can be from fractions of minutes to just a few minutes. Similarly in the pilot plant or factory, when the size of the crystallizer is up to hundreds or thousands of liters, the mixing scale lies in the macroscale domain, and the mixing time can take a few minutes or longer. It should be clearly noted that the definition of mixing scale and corresponding mixing time in this manuscript is based upon the absolute proportion of size. These definitions are different from the traditional definitions of micro-mixing, mesomixing, and macro-mixing which are based upon the mixing mechanisms and relative proportion of size.26 We propose to use the absolute proportion definitions here since it is more direct to describe the impact of scale on the mixing time. Also, the absolute proportion definition remains consistent with the traditional relative proportion definition in describing the microscale mixing, the mesoscale mixing, and the macroscale mixing in a macroscale crystallizer. The absolute proportion definitions help to clarify the confusion in ranking the mixing times between the “macro-mixing” of a 100 mL crystallizer which could be on the order of seconds, and the “meso-mixing” of a 4000 L industrial crystallizer which could be on the order of minutes, if using the traditional relative proportion definitions. By understanding the impact of scale on mixing time, it should be obvious that a constant addition time strategy for antisolvent or reactive crystallization processes may not be adequate for scale-up since mixing time will always increase. One remedy to compensate for the increase of mixing time upon scale-up is to increase the addition time (or decrease the addition rate) upon scale-up.30 Alternatively, it is feasible to apply a periodic pulse feed approach with the same overall addition time for scale-down evaluation. Upon pulsing, mixing time (in the laboratory crystallizer) will increase, thus mimicking the increase of mixing time upon scale-up. Other approaches include adjusting the agitator speed or using the subsurface addition tube near the agitator (see section 4.4). All these methods can help improve the mixing time upon scale-up. 4.2. Mixing Intensity: Scale-Up. For mixing intensity, it is well established that not all mixing intensity indexes, for example agitator tip speed or average power per unit volume, can be maintained upon scale-up. Figure 9 illustrates the requirement of agitator rate (rpm) at different scales under constant tip speed and constant power/volume criteria. Using the 1000 L size crystallizer at 100 rpm as reference, the agitator rpm for the 100 mL crystallizer would be 2000 or 800 rpm, respectively, on the basis of the constant tip speed rule or constant power/ volume rule. Either value would create significant vortexing and splashing of slurry in the 100 mL laboratory crystallizer and simply cannot be implemented in practice. It may not obvious at the first glance, but it is physically impossible to match both the mixing intensity and the mixing time upon scale-up of a geometrically similar mixer even under a uniform mixing environment. Referring back to Figure 8, in order to match the mixing time upon scale-up, it is necessary to increase the mixing intensity. Equivalently, if the mixing intensity is kept constant upon scale-up, the mixing time will be longer upon scale-up. In practice, this issue is addressed by

the particle breakage mechanism may be different. Applying this curve to cases using a rotor/stator homogenizer or stirred tank agitators is a very powerful correlation for wet milling of seed because it enables us to dial in a specific particle size of interest by selecting the proper fluid velocity with minimum trial-and-error. Another practical implication from this correlation can also be drawn near the region around 3−7 m/s. By extrapolating the curve into this region, it appears that the corresponding limiting particle size is ∼100 μm. Since the nominal agitator tip speed of pilot-plant or manufacturing stirred tanks falls into this region, as a corollary particles with sizes greater than ∼100 μm would likely be subjected to particle breakage during the crystallization. Consequently, it is recommended not to grow crystals beyond this size limit in order to avoid the complication of particle breakage in the pilot plant or commercial vessels. The impact of mixing intensity on secondary nucleation has been described with eq 2. The impact of mixing intensity on the primary nucleation can be found in references 26 and 28. The primary and the secondary nucleation mechanisms apply directly to the case of in situ seed generation approach. Despite the abundance of nucleation mechanisms in the literature,29 the practical approach still relies upon empirical correlations, and results are case specific. Further advancement to provide fundamental generalization among various cases and conditions would be valuable to the industry.

4. LEVEL III: PROCESS DYNAMICS (SCALE-UP) AND DESIGN OF CRYSTALLIZER At this level, we address the overall process dynamics (scale-up) in an environment with nonuniform suspension of slurry. The items of discussion include scale-up impact on mixing time, mixing intensity, and mixing distribution. Different types of crystallizers which are designed for meeting different mixing requirements will also be addressed in this section. 4.1. Mixing Time: Scale-Up. It has been long recognized that the scale has a profound impact on the mixing time. To illustrate it further, let us define the scale as below: • Microscale - from microliters to milliliters • Mesoscale - from milliliters to liters • Macroscale - from liters to thousands of liters On the basis of these definitions, Figure 8 is drawn from the author’s observations in practice to illustrate the qualitative

Figure 8. Relationship between mixing scale and mixing time. G

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

Figure 11. Comparison of local mixing intensity of completely full (left) and one-third full (right) 2000 L vessel under constant tip speed.

diagnose the mixing issues, beyond the traditional mixing rules or empirical correlations by constant tip speed or power per unit volume. However, extensive efforts are required initially to build the database of different vessels and agitators into the CFD. 4.4. Stirred Tank Crystallizer and External Recirculation Loop. Stirred tank vessels are the workhorses of pharmaceutical industry. The versatility of the tank in a large variety of chemical environments has made it the crystallizer of choice. These crystallizers range in size up to 20,000 L or more. A typical stirred tank crystallizer includes dual, down-pumping, pitched blade turbines or hydrofoils with a “tickler” impeller at the base, a subsurface addition line, baffles, and ram-type bottom outlet valve to aid in discharge of slurries. The ratio of agitator diameter to the vessel diameter could range from 0.4 to 0.5. In order to address the various mixing issues within the stirred tank upon scale-up, two methods are typically applied. The first is to vary (mostly increase) the agitator tip speed, and the second is to use the subsurface addition lines which are located at a region closely to the tip of the agitator. Both techniques improve the mixing time and increase the mixing intensity. As mentioned early, another means to address the nonuniform mixing environment is the use of an external recirculation loop shown in Figure 5. Such a design has been employed to study the impact of micromixing (feed line and in-line mixer), meso-mixing (recirculation loop), and macro-mixing (stirred tank crystallizer) independently.32 Given its flexibility in controlling the local and overall mixing environment, along with its ease of implementation with the existing facility, this scheme is gaining a wider acceptance in the pharmaceutical industry, in particular if wet-milled seed or in situ generated seed is used for the crystallization process. Since seed can be generated under a controlled mixing and supersaturation environment, this scheme also appears to be a much better candidate than the traditional stirred tank vessel for the semicontinuous or continuous mode of crystallization operations if desirable. 4.5. Alternatives to Stirred Tank Crystallizer. There are other types of crystallization equipment than a stirred tank crystallizer. One type is to produce more gentle mixing conditions than that needed by an impeller if solids are to be suspended. The other type is much more intense mixing where high levels of supersaturation are involved and nucleation is to be controlled. The former leads to the use of the fluidized bed1 (or equivalent) with focus on reducing the mixing intensity while maintaining the mixing distribution. The latter leads to the use of line-mixers such as a high speed rotor/stator homogenizer19 or impinging jet mixer,1,20,21 with focus on reducing the mixing time on the microscale at the feeding point and the controlled mixing at the downstream mesoscale environment.

Figure 9. Relationship between agitator rate RPM and crystallizer volume.

designing a crystallization process which is not sensitive to mixing intensity, such as via a proper choice of seed amount and size to avoid particle breakage and nucleation, a low supersaturation to minimize the secondary nucleation, and a proper choice of agitator to minimize the high local shear for nucleation, etc. Another strategy upon scale-up is to maintain the minimum agitation suspension of slurry, which can effectively reduce the degree of secondary nucleation as well.1,26 There will be more discussion in section 4.4. 4.3. Mixing Distribution. The most unique mixing feature upon scale-up is mixing distribution. When the batch size is small, it is reasonable to assume that the entire batch is wellmixed with similar mixing intensity throughout the entire crystallizer. Using computational fluid dynamics (CFD) simulation software, Figure 10 compares the local mixing intensity and

Figure 10. Comparison of local mixing intensity and micromixing time between the 250 mL (left) and the 2000 L (right) vessels under the constant power/volume criterion.

“micro-mixing time” (based upon the typical relative mixing definition) between a 250 mL crystallizer and 2000 L crystallizer31 under the constant overall power per volume criterion. As evident by the color and its distribution, the local mixing intensity and micromixing time are more uniform in the 250 mL crystallizer than those in the 500 gallon crystallizer. The vessel fill volume can affect the mixing distribution31 as well. Under the constant tip speed criterion for the identical 2000 L crystallizer, Figure 11 illustrates that the local mixing intensity and micromixing time are more uniform for the onethird-filled scenario than those of the completely filled scenario. CFD results are reliable in predicting the hydrodynamic behaviors of a single liquid phase system. As is evident in Figures 10 and 11, it can be highly effective to highlight and H

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

As shown in Figure 12, for fluidized bed operations, the solution enters at the bottom of the crystallizer and the supersaturation

5. CONCLUSIONS Industrial practice of crystallization of pharmaceuticals has been reviewed through three levels of perspectives. At the first level, thermodynamic properties, which include basic properties such as solubility, crystal form, and crystal morphology are addressed. At the second level, the transport properties and rate behavior, which include seed, control of supersaturation, mixing time, and mixing intensity, are addressed. At the third level, the overall process dynamics, which include mixing distribution and crystallizer design, are addressed.



AUTHOR INFORMATION

Corresponding Author

*[email protected] Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS The author sincerely thanks Drs. E. Paul, M. Midler, and J. McCauley for their constructive guidance and selfless support over the years in developing the crystallization practice. The author also thanks Dr. M. Diwan for his works on the development of in situ seed generation methodology, and Drs. N. Nere and A. Czyzewski for their works on the computational fluid dynamic simulation of stirred vessels.



Figure 12. Generic diagrams of impinging jet crystallizer (left) and fluidized bed crystallizer (right).

REFERENCES

(1) Tung, H.-H.; Paul, E.; Midler, M.; McCauley, J. Crystallization of Pharmaceuticals: An Industrial Perspective; Wiley: Hoboken, NJ, 2009. (2) Mullin, J. W. Crystallization, 4th ed.; Butterworth-Heinemann: Oxford, 2001. (3) Laird, T. Org. Process Res. Dev. 2009, 13, 1214. (4) Laird, T. Org. Process Res. Dev. 2005, 9, 857. (5) Laird, T. Org. Process Res. Dev. 2003, 7, 957. (6) Laird, T. Org. Process Res. Dev. 2000, 4, 370. (7) Chen, C.-C.; Song, Y. Ind. Eng. Chem. Res. 2004, 43, 8354−8362. (8) Tung, H.-H.; Tabora, J.; Variankaval, N.; Bakken, D.; Chen, C.-C. J. Pharm. Sci. 2008, 97, 1813−1820. (9) Bernstein, J. Cryst. Growth Des. 2011, 11, 632−650. (10) Chemburkar, S. R.; Bauer, J.; Deming, K.; Spiwek, H.; Patel, K.; Morris, J.; Henry, R.; Spanton, S.; Dziki, W.; Porter, W.; Quick, J.; Bauer, P.; Donaubauer, J.; Narayanan, B. A.; Soldani, M.; Riley, D.; McFarland, K. Org. Process Res. Dev. 2000, 4, 413−417. (11) Storey, R.; Ymen, I. Solid State Characterization of Pharmaceuticals; Wiley: West Sussex, UK, 2011. (12) Singh, M. R.; Chakraborty, J.; Nere, N.; Tung, H.-H.; Bordawekar, S.; Ramkrisha, D. Cryst. Growth Des. 2012, 12, 3735− 3748. (13) Sheth, A. R.; Grant, D. J. W. KONA 2005, 23, 36−47. (14) Lovett, M. A.; Muratore, M.; Doherty, M. AIChE J. 2011, 58, 1465−1474. (15) Randolph, A. D.; Larson, M. A. Theory of Particulate Processes, 2nd ed.; Academic Press: San Diego, CA, 1988. (16) Beckmann, W. Org. Process Res. Dev. 2000, 4, 372−383. (17) Johnson, B. K.; Tung, H.-H.; Lee, I.; Cote, A. S.; Starbuck, C.; Midler, M. U.S. Patent Appl. 2009 0087492, 2009. (18) Diwan, M.; Tung, H.-H.; Kim, E.; Bordawekar, S. In-situ Seed Generation with Wet Milling/Energy for Pharmaceutical Crystallization. AIChE Annual Meeting, Salt Lake City, UT, 2010. (19) Kamahara, T.; Takasuga, M.; Tung, H.-H.; Hanaki, K.; Fukunaka, T.; Izzo, B.; Nakada, J.; Yabuki, Y.; Kato, Y. Org. Process Res. Dev. 2007, 11, 699−703. (20) Midler, M.; Paul, E. L.; Whittington, E. F.; Futran, M.; Liu, P. D.; Hsu, J.; Pan, S. H. U.S. Patent 5,314,506, 1994. (21) Johnson, B.; Prud’homme, R. L. AIChE J. 2003, 49, 2264−2282.

or portion of it is released as the supernatant flows out from the top of fluidized bed. The slurry is suspended by the upward flow. Similar to a plug flow reactor, fluidized bed crystallizers can exhibit a good degree of lateral homogeneity with a very low degree of back mixing in comparison to a stirred tank crystallizer. Due to its low fluidizing velocity and mixing intensity, it minimizes nucleation and can achieve growth at low supersaturation. Additionally, classification of particle size can be accomplished along the height of fluidized bed of varying diameter. It is primarily used in continuous or semicontinuous modes of operation. Similar to the stirred tank crystallizer with an external recirculation loop, the fluidized bed crystallizer can be equipped with an external recirculation loop containing a wet mill for the generation of wet seed. For in-line mixers such as impinging jet mixers, it is at the opposite end of the spectrum with a much higher mixing intensity. They are microscale devices with very high mixing intensity as streams of reactants or an antisolvent and a solution impinge at high velocity, leading to a rapid localized mixing. As a result, the mixing time is shorter than the induction time. Even at a level of supersaturation beyond the metastable limit, it allows nucleationdominated processes under uniform conditions throughout the operation. In general, they produce finer particle sizes often on the order of a few micrometers or less. It is primarily used in continuous or semicontinuous mode of operations. Similar to the stirred tank crystallizer with an external loop, the in-line mixer can be coupled with an external circulation loop for controlling the downstream mesoscale mixing beyond the feeding point. Both the fluidized bed crystallizer and the impinging jet crystallizer have been successfully implemented in commercial production of pharmaceuticals.1 The readers can find more detailed discussion in the cited reference. I

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development

Review

(22) Davey, R. J.; Blagden, N.; Righini, S.; Alison, H.; Ferrari, E. S. Phys. Chem. B 2002, 106, 1954−1959. (23) Togkalidou, T.; Tung, H.-H.; Sun, Y.; Andrews, A.; Braatz, R. D. Ind. Eng. Chem. Res. 2004, 43, 6168−6181. (24) Zhou, G. X.; Fujuwara., M.; Woo, X. Y.; Rusli, E.; Tung, H.-H.; Starbuck, C.; Davidson, O. A.; Ge, Z.; Braatz, R. D. Cryst. Growth Des. 2006, 6, 892−898. (25) Nagy, Z. K.; Braatz, R. D. Annu. Rev. Chem. Biomol. Eng. 2012, 3, 55−75. (26) Paul, E. L.; Atiemo-Obeng, V.; Kresta, S. M. Handbook of Industrial Mixing: Science and Practice; Wiley: New York, 2003. (27) Tung, H.-H.; Robertson, S.; Felmet, K. ; Starbuck, C. ; Tom, J. An Evaluation of Cavitation Milling to Achieve Particle Size Reduction of Active Pharmaceutical Ingredients. AIChE Annual Meeting, Reno, NV, 2001. (28) O’Grady, O.; Barrett, M.; Casey, E.; Glennon, G. Trans IChemE, Part A, Chem. Eng. Res. Des. 2007, 85, 945−952. (29) Mersmann, A. Crystallization Technology Handbook, 2nd ed.; Marcel Dekker: New York, 2001. (30) Liu, X.; Hatziavramidis, D.; Arastoopour, H.; Myerson, A. AIChE J. 2006, 52, 3621−3625. (31) Czyzewski, A.; Tung, H.-H. ; Bordawekar, S. V.; Nere, N. K. Harness Computational Fluid Dynamics (CFD) to Guide Reliable Scale-up of Pharmaceutical Processes. AIChE Annual Meeting, Minneapolis, MN, 2011. (32) Torbacke, M.; Rasmussen, A. C. Chem. Eng. Sci. 2001, 56, 2459−2473.

J

dx.doi.org/10.1021/op3002323 | Org. Process Res. Dev. XXXX, XXX, XXX−XXX