Advanced Drug Delivery Reviews 56 (2004) 275 – 300
`
`www.elsevier.com/locate/addr
`
`High-throughput crystallization: polymorphs, salts, co-crystals
`and solvates of pharmaceutical solids
`
`Sherry L. Morissette a,*, O¨ rn Almarsson a, Matthew L. Peterson a,
`Julius F. Remenar a, Michael J. Read a, Anthony V. Lemmo a, Steve Ellis a,
`Michael J. Cima b, Colin R. Gardner a
`
`a TransForm Pharmaceuticals, Inc., 29 Hartwell Avenue, Lexington, MA 02421, USA
`b Department of Materials Science and Engineering, Massachusetts Institute of Technology, Room 12-011, 77 Massachusetts Avenue,
`Cambridge, MA 02139, USA
`
`Received 21 August 2003; accepted 6 October 2003
`
`Abstract
`
`The concepts of high-throughput (HT) screening and combinatorial synthesis have been integrated into the pharmaceutical
`discovery process, but are not yet commonplace in the pharmaceutical development arena. Emerging strategies to speed
`pharmaceutical development and capture solid form diversity of pharmaceutical substances have resulted in the emergence of
`HT crystallization technologies. The primary type of diversity often refers to polymorphs, which are different crystal forms of
`the same chemical composition. However, diverse salt forms, co-crystals, hydrates and solvates are also amenable to study in
`HT crystallization systems. The impact of form diversity encompasses issues of stability and bioavailability, as well as
`development considerations such as process definition, formulation design, patent protection and regulatory control. This
`review highlights the opportunities and challenges of HT crystallization technologies as they apply to pharmaceutical research
`and development.
`D 2003 Elsevier B.V. All rights reserved.
`
`Keywords: High-throughput; Crystallization; Polymorph; Solvate; Salt; Co-crystal
`
`Contents
`
`1.
`2.
`3.
`
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Introduction .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Development of high-throughput crystallization technologies .
`Applications of high-throughput crystallization screening in pharmaceutical research and development: case studies.
`3.1.
`High-throughput salt selection .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`3.2.
`Solid form discovery in highly polymorphic systems .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`3.3.
`Avoiding latent polymorphism .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`3.4.
`Prediction of crystallization and polymorphism: applications to pharmaceutical form studies .
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`
`276
`278
`285
`285
`287
`290
`291
`
`* Corresponding author. Tel.: +1-781-674-7823; fax: +1-781-863-6519.
`E-mail address: morissette@transformpharma.com (S.L. Morissette).
`
`0169-409X/$ - see front matter D 2003 Elsevier B.V. All rights reserved.
`doi:10.1016/j.addr.2003.10.020
`
`Gilead 2015
`I-MAK v. Gilead
`IPR2018-00126
`
`

`

`276
`
`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`.
`.
`.
`.
`.
`Engineering of co-crystals
`3.5.
`Post-screening analyses and form selection .
`4.
`Summary and outlook .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`5.
`References .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`292
`295
`296
`297
`
`1. Introduction
`
`Active pharmaceutical ingredients (APIs) are fre-
`quently delivered to the patient in the solid-state as part
`of an approved dosage form (e.g., tablets, capsules,
`etc.). Solids provide a convenient, compact and gen-
`erally stable format to store an API or a drug product.
`Understanding and controlling the solid-state chemis-
`try of APIs, both as pure drug substances and in
`formulated products, is therefore an important aspect
`of the drug development process. APIs can exist in a
`variety of distinct solid forms, including polymorphs,
`solvates, hydrates, salts, co-crystals and amorphous
`solids. Each form displays unique physicochemical
`properties that can profoundly influence the bioavail-
`ability, manufacturability purification, stability and
`other performance characteristics of the drug [1].
`Hence, it is critical to understand the relationship
`between the particular solid form of a compound and
`its functional properties. Discovery and characteriza-
`tion of the diversity of solid forms of a drug substance
`provide options from which to select a form that
`exhibits the appropriate balance of critical properties
`for development into the drug product. Importantly,
`the desired properties may vary with each mode of
`delivery (i.e., oral, pulmonary, parenteral, transdermal,
`etc.), such that the solid form may differ for each
`optimized dosage form. Given these options, the
`choice and design of pharmaceutical solid forms can
`be critically important to successful drug development.
`Solid form discovery and design depends on the
`nature of the molecule of interest and type of physical
`property challenges faced in its development. The
`preferred solid form is generally the thermodynami-
`cally most stable crystalline form of the compound
`[1,2]. However, the stable crystal form of the parent
`compound may exhibit inadequate solubility or dis-
`solution rate resulting in poor oral absorption, partic-
`ularly for water-insoluble compounds. In this case,
`alternative solid forms may be investigated. For
`ionizable compounds, preparation of salt forms using
`pharmaceutically acceptable acids and bases is a
`common strategy to improve bioavailability [1,3,4].
`
`Like the parent compound, pharmaceutical salts may
`exist in several polymorphic, solvated and/or hydrated
`forms.
`Most APIs and their salts are purified and isolated
`by crystallization from an appropriate solvent during
`the final step in the synthetic process. A large number
`of factors can influence crystal nucleation and growth
`during this process, including the composition of the
`crystallization medium and the process(es) used to
`generate supersaturation and promote crystallization
`[1,5 – 13]. The most notable variables of composition
`and processing are summarized in Table 1. Solid form
`screening is used to understand the effects that these
`variables have on the polymorphic outcome of a
`crystallization experiment, so that a robust process
`can be identified to produce the desired crystal form.
`Traditionally,
`the study of solid form diversity of
`active compounds has relied on the use of a variety
`of common process methods for generation of new
`forms, coupled with modern characterization methods
`for analysis of the solids produced [2,14]. Most often,
`however, a combination of solvent recrystallization
`(cooling or evaporative, as well as slurry conversion)
`and thermal analysis (e.g., hot stage microscopy,
`differential scanning calorimetry) are employed for
`initial form screening. Such methods are inherently
`slow and only allow exploration of a small fraction of
`the composition and process space that can contribute
`to form diversity. Before suggesting a form for devel-
`opment, scientists may have carried out only a few
`dozen crystallization experiments and possibly pre-
`pared a handful of different salts of a compound. The
`main reasons for the limited number of experiments
`are the constraints on availability of compound and
`scientists’ analytical capacity in a given time frame,
`and they are therefore often forced to make form
`selection decisions on incomplete data. Accordingly,
`it
`is not surprising that unexpected and undesired
`outcomes can, and do, occur later on in development.
`Despite more than a century of research [15], the
`fundamental mechanisms and molecular properties
`that drive crystal form diversity, specifically the
`nucleation of polymorphic forms, are not well under-
`
`

`

`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`277
`
`Table 1
`Crystallization composition and processing variables [1,2,8]
`Process variablesa
`
`Composition type
`
`Thermal
`
`Anti-solvent
`
`Evaporation
`
`Slurry conversion
`
`Other variables
`
`n Heating rate
`
`n Anti-solvent
`type
`
`n Rate of
`evaporation
`
`n Solvent type
`
`n Mixing rate
`
`n Cooling rate
`
`n Maximum
`temperature
`
`n Rate of anti-
`solvent addition
`n Temperature
`of anti-solvent
`addition
`
`n Evaporation time
`
`n Carrier gas
`
`n Incubation
`temperature
`n Incubation
`time
`
`n Incubation
`temperature(s)
`n Incubation time
`
`n Time of anti-
`solvent addition
`
`n Surface-volume
`ratio
`
`n Thermal cycling
`and gradients
`
`n Impeller design
`
`n Crystallization
`vessel design
`(including
`capillaries, etc.)
`
`Polymorph/
`solvates
`
`n Solvent/
`solvent
`combinations
`n Degree of
`supersaturation
`n Additive type
`
`Salts/
`co-crystals
`
`n Counter-ion
`type
`
`n Acid/base
`ratio
`n Solvent/
`solvent
`combinations
`
`n Additive
`concentration
`
`n Degree of
`super-saturation
`n Additive type
`and concentration
`n pH
`n Ionic strength
`a Applicable to all types of screens.
`
`stood [13,16]. As a result, predictive methods of
`assessing polymorphic behavior of pharmaceutical
`compounds by ab initio calculations remain a formi-
`dable challenge. Even in cases where the existence of
`a crystalline form is predicted, the stability relative to
`other crystalline packing arrangements has been dif-
`ficult to estimate with accuracy [17]. Moreover, the
`prediction of packing structures for multicomponent
`(e.g., solvates, hydrates, co-crystals) or ionic systems
`is not yet possible [17]. Due to these limitations, solid
`form discovery remains an experimental exercise,
`where manual screening methods are employed to
`explore form diversity of a compound.
`the drug
`Control over solid form throughout
`development process is of paramount
`importance.
`Reliable preparation and preservation of the desired
`form of the drug substance must be demonstrated,
`and has become increasingly scrutinized by regula-
`tory agencies as more sensitive and quantitative
`solid-state analytical methods have become available
`[18]. Many strategies to influence and control the
`crystallization process to produce the solid form of
`interest have been reported. Some examples include
`stereochemical control using tailor-made auxiliaries
`[19 – 21], targeted solvent recrystallization [22 – 24],
`and templating using a variety of surfaces (e.g.,
`organic single crystal substrates [25], surfaces of
`metastable crystal faces [25,26],
`inorganic crystal
`
`surfaces [27] and polymeric materials [28]). Recent
`studies have also begun to uncover the role of
`reaction byproducts and other impurities in determin-
`ing polymorphic outcome and crystal properties
`[29 – 32], and in fact, it has been shown that in some
`cases such species can stabilize metastable crystal
`forms [33,34]. In addition, new processing methods
`continue to be developed to improve discovery and
`characterization of new forms, including precipitation
`by supercritical fluid [35,36],
`laser induced nucle-
`ation [37 – 39] and capillary crystallization [40 – 42].
`However, there remains a lack of fundamental un-
`derstanding of the nucleation process and the specific
`factors that contribute to crystallization of diverse
`forms of a compound [13,21,23]. In order to fully
`control the crystallization process, the link between
`the physical or chemical processes that
`influence
`nucleation and crystal growth needs to be better
`established. It is in this area that new experimental
`methodologies have the potential to enable develop-
`ment of this knowledge base.
`There is reason to believe that the already compli-
`cated landscape of pharmaceutical solid forms will
`become even more complex in the future. It is now
`increasingly appreciated that hydrogen bonded co-
`crystal structures between active agents and molecules
`other than water or solvent can be prepared. For
`example, co-crystals of aspirin, rac-ibuprofen and
`
`

`

`278
`
`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`rac-flurbiprofen have been prepared by disrupting the
`carboxylic acid dimers using 4,4V-bipyridine [43].
`These structures are formally molecular compounds
`(or co-crystals) but do not
`involve formation of
`covalent bonds or charge transfer from or to the active
`substance. Recent demonstrations of these principles
`with drug compounds have been published [43 – 45].
`Exploration of a given compound’s polymorphs,
`hydrates, solvates, salts, co-crystals and combinations
`of all of these appears intractable by conventional
`experimental methods, and as the number of potential
`methods for exploring and controlling crystal form
`diversity continue to expand, existing strategies will
`become increasingly inadequate. In an effort to un-
`derstand form diversity in a more comprehensive
`manner, high-throughput (HT) crystallization systems
`have recently been developed. This methodology uses
`a combinatorial approach to solid form generation,
`where large arrays of conditions and compositions are
`processed in parallel. Experiments are performed at
`small scale to reduce the material demand and to
`afford the largest number of conditions possible.
`The large number of crystallization trials performed
`in these experiments reflects the reality that nucleation
`rate has an extremely non-linear dependence on the
`experimental conditions, and as such, the probability
`of a chance occurrence of a particular form is in-
`creased by a HT approach. Supersaturation (solubility)
`and induction time of the various possible solid forms
`are independently controlled by these conditions,
`resulting in highly non-linear time dependence of
`crystallization. In addition,
`the combinatorial ap-
`proach permits exploration of a chemical continuum,
`where use of many solvent mixtures may allow one to
`assess what underlying physical or chemical processes
`are required to produce a particular solid form. Once a
`variety of conditions that can be used to produce a
`given crystal form on the microscale are identified in
`the HT screen, scale-up studies are typically con-
`ducted to optimize the process for laboratory scale
`production.
`In this review, the development and application of
`novel HT crystallization technologies for exploration
`of solid form diversity are discussed. The operational
`features of a fully integrated, automated HT crystal-
`lization system are presented, highlighting the design
`requirements for hardware and software components,
`as well as general specifications for consumables.
`
`Case studies are used to illustrate the benefits and
`capabilities of the approach, including salt selection in
`early lead optimization (ELO) and pre-clinical devel-
`opment, polymorph and solvate screening in highly
`polymorphic systems, comprehensive discovery of
`crystal forms to reduce the risk of late displays of
`polymorphism, comparison of experimental and pre-
`dictive methods of solid form discovery, and engi-
`neering of co-crystals. The need for post-screening
`characterization of crystal forms to enable ranking and
`selection of the most suitable form for development is
`briefly reviewed. Finally,
`the implications of HT
`crystallization technologies on the future of solid form
`screening processes, intellectual property protection
`and regulatory compliance are discussed.
`
`2. Development of high-throughput crystallization
`technologies
`
`HT crystallization systems have been developed to
`more rapidly and comprehensively explore the multi-
`parameter space that contributes to solid form diver-
`sity [40,46 – 51]. In its simplest description, HT
`crystallization can be broken down into three key
`experimental steps: design of experiment (DOE),
`execution of experimental protocols and analysis of
`data. Systems designed to carry out these experiments
`generally consist of both hardware and software
`components that drive and track experimentation,
`and permit data storage, retrieval and analysis. Such
`systems should be designed to be flexible and scalable
`to ensure that a variety of experimental procedures
`can be carried out either serially or concurrently.
`Thus, the system can be employed at various stages
`of drug development, where differences exist in the
`quality and quantity of compound available. While it
`is highly desirable to have the ability to mine and
`model experimental data, and to use the subsequent
`knowledge to guide further experiments, not all HT
`crystallization systems are equipped with these fea-
`tures. In Section 3, the hardware and software con-
`siderations for design and development of a fully
`integrated, informatics-driven HT crystallization sys-
`tem are described.
`While the concepts of HT screening are widely
`applied in the pharmaceutical industry, most notably
`in the drug discovery arena [52], the application of
`
`

`

`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`279
`
`in particular
`HT approaches to drug development,
`solid form screening, are just beginning to be real-
`ized. These latter approaches, however, are more akin
`to HT experimentation than HT screening. Hence,
`several important distinctions, which reflect on the
`design of HT experimental systems, need to be made.
`First,
`the goal of HT screening is to get a small
`number of successful outcomes, which are then
`passed on to the next stage of development. Little
`effort is typically made to learn why certain outcomes
`were positive and why others were negative. In
`contrast, HT experimentation, such as HT crystalliza-
`tion, is carried out with the goal of having each point
`in the experiment produce multiple types of data that
`can be interpreted, and the interpretation used to
`guide the experimental process to a successful con-
`clusion. Second, unlike traditional HT screening
`assays where experiments are generally conducted
`under constant experimental conditions, HT crystalli-
`zation experiments for solid form discovery are best
`conducted using a variety of process methods, each
`having varying experimental conditions (e.g., temper-
`ature variations as a function of time) over the course
`of the experiment. These additional process variables
`permit maximal diversity in the experimental space,
`increasing the likelihood that comprehensive cover-
`age will be achieved. Finally, there is a distinction to
`be made in terms of relative ‘‘hit rates’’. In both HT
`screening and HT crystallization, a ‘‘hit’’ can be
`
`thought of as a set of conditions that gives rise to a
`desired result. In HT screening, the desired result is
`typically an activity, or potency, that exceeds a pre-
`defined threshold. In HT crystallization, a hit
`is
`defined as the formation of a solid. The typical
`observed hit rate of HT screening is on the order of
`0.1% of the total number of samples analyzed. In
`contrast, HT crystallization experiments can yield hit
`rates ranging from tens of percents to nearly 100%,
`depending on the type of experiment and the process
`mode(s) used. For example, while only a handful of
`compounds from a selection of thousands may exhibit
`the required potency, 10 – 50% of crystallization trials
`may yield solids. In fact, the range of wells that yield
`solids is very wide, depending on process mode and
`experimental
`time scale, as will be discussed in
`subsequent sections. The impact of these differences
`is manifested in the design and operational require-
`ments of HT experimentation systems.
`A fully integrated HT crystallization system con-
`sists of a number of components, including experi-
`mental design and execution software, robotic
`dispensing and handling hardware, automated high-
`speed micro-analytical tools, end-to-end sample track-
`ing and integrated cheminformatics analysis software
`for data visualization, modeling and mining. A sche-
`matic overview detailing the workflow of such a
`system is depicted in Scheme 1 [53]. These features
`are supported by a comprehensive informatics foun-
`
`Scheme 1. A schematic illustration of the workflow of a fully integrated HT crystallization system [53].
`
`

`

`280
`
`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`dation that is used to handle the large quantities of
`data generated. Specifically, informatics tools are used
`to design statistically relevant and diverse experi-
`ments, drive the automation hardware to perform the
`specified operations, and provide an analytical func-
`tion to analyze, compare and sort
`the results of
`experiments. An important feature of these systems
`is the ability to mine and model experimental data and
`use the knowledge generated to guide further experi-
`ments. These functions are supported by use of a
`relational database that provides a mechanism of
`communication between system components.
`When designing a HT crystallization experiment, or
`set of experiments, a large variety of parameters of
`composition and process are involved. Experimental
`designs must be aimed at covering a large multifacto-
`rial parameter space, with the goal of determining
`which experimental factors affect the desired outcome.
`In practice, it is desirable to place constraints on the
`experimental space, making common statistical design
`methods such as full or partial factorial designs inap-
`propriate or impractical. For example, hardware limi-
`tations, including minimum and maximum dispense
`volumes or masses and accessible temperature ranges,
`as well as constraints related to chemical compatibility
`(i.e., reactivity of components, miscibility, etc.) or
`toxicity limits of components (if appropriate), need
`to be considered. Thus, alternative DOE methods that
`can accommodate such constraints are required. D-
`optimal design [54,55] is an example of a DOE
`algorithm that can take a set of constraints, such as
`the ones described above, in combination with a target
`analytical model and determine the optimal set of
`experimental points to test. Another commonly used
`DOE algorithm is diversity generation, with which the
`experimentalist selects a set of pertinent chemical
`properties and uses the algorithm to evenly spread
`experimental points over the chosen property space. In
`addition, some systems utilize a solubility calculator
`tool to estimate the solubility of the API in the given
`solvent/additive mixture. The calculated information is
`then used to select the appropriate concentration of
`API in each mixture so that it is supersaturated with
`respect to the reference phase at the harvest tempera-
`ture. Here, the driving force for crystallization can also
`be varied by tailoring the composition of each sample
`based on the API solubility in that mixture. With such
`DOE tools, experiments may be designed to effective-
`
`ly and simultaneously explore the diverse composition
`and process space described in Table 1.
`Ideally, DOE algorithms should also incorporate
`prior knowledge or experimental results, which have
`been stored in a database as a set of rules or models, to
`limit an experimental space to have certain predicted
`characteristics. For example, over the course of time, a
`regression model may be developed between a set of
`known or calculated chemical properties and a pa-
`rameter of experimental interest. The model could be
`used during the design of a new experiment in order to
`test only those chemicals that are predicted to give a
`desirable result. Since a large number of factors need
`to be considered during experimental design, the DOE
`interface available to the scientist must not only be
`flexible and easy to use, but must also offer tools that
`aid design efficiency and effectiveness and permit
`input of scientific knowledge generated over time.
`At the end of the experimental design process, the
`resulting set of experimental conditions is translated
`into a series of commands for the HT systems, and
`stored in a relational database for later retrieval by the
`software that controls the automation. When an ex-
`periment is activated,
`the overall operation of the
`automation systems is managed by the HT informatics
`system, which is responsible for physical operation of
`the HT platforms as well as data tracking and storage.
`Execution of experimental commands is carried out
`by automated laboratory equipment that comprises the
`HT crystallization system. Specialized automated sys-
`tems perform several of the functions in a sequence of
`events that make up the experiment. Each station is
`controlled through an interface to the informatics
`system that ensures the samples are processed at the
`correct stations, in the correct order, with the selected
`experimental parameters being followed. Parameters
`of operation are recorded, including the time at which
`an action is taken. After execution of the experimental
`steps, the software interface retrieves any pertinent
`information generated by the automated platform,
`such as assay results or operational parameters, stores
`these data in the relational database, and updates the
`status of the experiment to reflect the completion of
`operations.
`In general, the hardware required for a HT crystal-
`lization system is comprised of four major functional
`elements: sample preparation, solids generation, solids
`detection and sample analysis. Sample preparation
`
`

`

`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`281
`
`involves adding the compound of interest (API) to the
`diverse set of conditions used to conduct crystalliza-
`tion studies. Typically,
`the API is dispensed as a
`solution in a suitable solvent, followed by solvent
`removal to yield the solid API. Solvent removal can
`be achieved by passive evaporation or by controlled
`active evaporation (e.g., use of a vortex dryer).
`Alternatively, the API can be delivered in the solid
`state with suitable powder handling systems. Depend-
`ing on the amount of saturation desired, the crystal-
`lization vessel used, and the API’s solubility in
`solvents or solvent mixtures of interest, API masses
`ranging from a few hundreds of micrograms to several
`milligrams will be present in each vessel. Once the
`API has been delivered to the crystallization vessels
`(tubes, vials or microwell plates), combinations of
`solvents and/or additives are added to each vessel. By
`taking advantage of the power of combinatorial
`approaches, large numbers of unique combinations
`can be dispensed from manageable sets of starting
`materials.
`Compatibility of equipment components (syringes,
`dispense tips, tubing, etc.) and consumables (plates,
`tubes, etc.) with solvents and other compounds is a
`key hurdle faced in the development of combinatorial
`crystallization for small molecules. Unlike protein
`crystallization systems [56,57], which are commonly
`based on the sitting-drop method in aqueous media,
`small molecule crystallization employs a range of
`crystallization additives and processes. The additives
`include organic solvents with varying properties (e.g.,
`alcohols, acetone, hexane, ethyl acetate, etc.), water,
`acids, bases and co-crystal formers, as well as other
`compounds (e.g., small molecule templating agents,
`surfactants, pharmaceutical excipients, etc.). This
`wide range of materials needs to be handled by
`appropriate liquid handling techniques to enable the
`combinatorial assembly previously mentioned. Ideal-
`ly, liquid transfers are achieved using multichannel
`pipettors with individually controllable channels.
`Depending on the crystallization vessel design, the
`volumes of reagents dispensed will be as low as a few
`microliters to as high as several hundred microliters.
`Potential for cross-contamination and tendency
`toward unwanted solvent evaporation from crystalli-
`zation wells are challenges that need to be addressed
`in a HT crystallization system. A large number of the
`solvents used to crystallize small molecules have high
`
`vapor pressure under ordinary laboratory conditions.
`Sealing of the crystallization vessels is key to being
`able to control composition during crystallization
`from these solvents. Due to solvent fugacity, vessels
`need to be protected from ingress of the components
`of neighboring wells. These problems have been
`solved by different means, such as sealing of individ-
`ual tubes with a Teflon-backed crimp seal [40] or O-
`rings/gasket seals and clamped covers [47,51].
`HT crystallization must enable several process
`modes that are compatible with the compound (e.g.,
`chemical stability,
`thermal stability, etc.). In some
`cases, multiple modes of operation may be combined.
`The most common modes of solids generation will be
`discussed below, including thermal cooling crystalli-
`zation, anti-solvent and evaporative crystallization.
`Less common process modes include melt crystalli-
`zation, flash or quench cooling and template-directed
`crystallization. It is important to note that generation
`of maximal diversity in solid form requires multiple
`modes of operation [6,18,58].
`In thermally induced cooling crystallization, sam-
`ples created in the sample preparation process de-
`scribed above are subjected to temperature ramps.
`Prior to beginning the temperature ramp, samples are
`exposed to an elevated temperature for a short period
`of time in order to dissolve the API in the crystalliza-
`tion medium. Although dissolution can be achieved
`most simply by diffusion and convection from the
`heating process, addition of external energy can speed
`up the process (e.g., sonication). Samples may be
`optically inspected (see Fig. 1) and vessels that contain
`undissolved solids can be flagged in the database for
`further analysis. For instance, undissolved samples
`may be treated as slurry conversion experiments and
`monitored over time for crystal form changes. The
`thermal cycle is then initiated, using controlled cooling
`to induce supersaturation. In this mode of crystalliza-
`tion, samples continually experience an under cooling
`and, based on the level of supersaturation in the vessel,
`may recrystallize at a given temperature after a period
`of time. Thermal crystallization tends to generate a
`cumulative number of samples that are produced over
`time in a fashion approximating a square root function,
`as illustrated in Fig. 2. This means that initially there is
`a small bolus of ‘‘hits’’, after which the rate of
`crystallization tails off over a period of time, typically
`in days to weeks. This results in a manageable hit rate
`
`

`

`282
`
`S.L. Morissette et al. / Advanced Drug Delivery Reviews 56 (2004) 275–300
`
`methods, differential rates of solvent loss from mix-
`tures result in unknown composition of the crystalli-
`zation medium at the time of crystal nucleation. In
`addition, the degree of supersaturation changes over
`the course of the experiment, often resulting in the
`appearance of multiple crystal forms. The evaporative
`mode of solids generation typically produces through-
`put and hit rates intermediate between the thermal and
`anti-solvent processes.
`As suggested above, in appropriately configured HT
`crystallization systems, several process modes may be
`used in series or in parallel [40]. Frequently, the
`preparation of replicate plates (in some systems
`‘‘daughter’’ plates [47,51]) is necessary for parallel
`processing by different process modes. Systems may
`be additionally equipped with the ability to serially
`process sample arrays using different process modes
`[59]. This feature is particularly attractive for cases
`where only small quantities of sample are available,
`increasing the drive to generate useful information
`from every sample. Here, samples may be processed
`by optimal modes first (e.g., thermal crystallization),
`then a secondary process step can be applied to max-
`imize the hit rate. Another example where this feature is
`useful is in the case of salt selection, especially in early
`drug discovery. Upon the addition of salt forming acids
`or bases, the solubility of the compound is modulated
`by in-situ salt formation, often resulting in reduced or
`non-existent driving forces for crystallization (e.g.,
`subsaturation) of the salt species, particularly in polar
`
`Fig. 2. Typical rate of appearance of solids during a thermally driven
`HT crystallization experiment [65].
`
`Fig. 1. Photo of optical inspection station. (Inset shows close up of
`crystallization vessel that contains crystals.) (Courtesy of Trans-
`Form Pharmaceuticals, 2002.)
`
`for analysis, on the order of approximately 10% in
`aggregate. This mode of solids generation has the
`lowest throughput rate, typically, because experiments
`span days to weeks, with system residence times of
`months being possible.
`In contrast, anti-solvent addition, also known as
`‘‘crash-out’’ (or ‘‘drown out’’) crystallization, relies
`on the fact that an API is soluble to varying degrees in
`the crystallization medium, but is largely insoluble in
`a particular solvent or solvents (e.g., the anti-solvent).
`As a result, this mode of crystallization can operate at
`high-throughput rates, with samples being turned
`arou