Quality by Design and Design Space
The concept of Quality by Design (QbD) is described in the ICH guideline Q8 (R1) [1]. The guideline, which is written for the development of the drug product, includes the following definition of QbD: a systematic approach to development that includes incorporation of prior knowledge, results of studies using design of experiments, use of quality risk management and use of knowledge risk management throughout the life cycle of the drug product. Since the introduction of ICH Q8, much progress has been made in introducing the QbD concept for the manufacturing process of the active pharmaceutical ingredient (API) as well. This article focuses on the final API crystallization.
Design of experiments and quality risk management, as part of process development, are not new but have been used in the pharmaceutical industry in a more limited way. Typically, in the late stage of a drug project, experiments called “method qualification (MQ)” or “estimation of parameter ranges” (EPR) have been performed. Normally, this work includes a series of experiments where the primary focus is to show that the process yields a consistent result- without any exceptions- within a certain process parameter range. The obtained parameter intervals, the traditional design space, are normally the only data that are submitted to the regulatory authorities. 
A consequence of this narrow filing strategy has been that even relatively small changes in the process must go through a post-approval process. When applying the new QbD principles and submitting a QbD-file, data and information are given on a wider “design space” as well as an even larger “knowledge space.” Schematically, this is illustrated in Figure 1. Notice that the knowledge space can be unsymmetrical and include areas (life cycle management (LCM) design space) which may be used in a future process. By submitting a wider range of data, which shows sensitivity and responses to changes in certain process parameters as well as describing a deeper understanding of the process, it will become easier to adjust and optimize the process to for instance new starting materials or changes in scale and type of equipment. Consequently, the regulatory flexibility is much larger.
What Does QbD for API’s Mean?
Dr M Nasr, Dir. ONDQA, FDA stated in 2005 a recommendation for how to proceed with QbD work not only for the drug product but also for the API’s [2]. He suggested the following: (i) design the drug substance with patient needs and drug product & process design in mind, (ii) Start by determining critical quality attributes (CQA:s) impacting safety and efficacy (e.g. impurity content), drug product performance (e.g. dissolution, stability) and manufacturability (e.g. content uniformity), (iii) For each unit operation, make sure you understand how process parameters affect the CQA, i.e. determine operating ranges of process parameters and input material attributes that achieve desired quality (the design space) and establish appropriate process controls to minimize effects of variability on CQA.
Quality Risk Assessment (QRA)
The quality risk assessment includes a step by step analysis of the process either by a topdown or bottom-up approach. The top-down approach aims to identify which process steps may have an influence on the CQA. This is useful in the early stages of process development. The bottom-down procedure, which should not be applied until the process is clearly defined, identifies possible failures in every process step and operation, and the consequences of the failure on the CQA are estimated. Typical examples of process failures are temperatures, concentrations, amounts or times that are out of the recommended ranges. By estimation of the severity, probability and detectability of the process failures, the most important ones are identified and the problem may be avoided by a change in design or by introducing necessary process controls.
Why Apply QbD Principles?
The application of QbD and quality risk assessment may initially mean more experimental and paper work, especially when the research and development organization is not used to the new way of working. However, it is clear that QbD has many advantages. First, and most important, it may give an improved regulatory flexibility with less post approval work. Eventually this will save both money and work. Secondly, it offers an effective tool for faster and more focused process development. Finally, as a consequence of the risk analysis performed, any non-robust process steps will be identified and possibly eliminated.
The Design of a Robust Crystallization Process
By introducing the above suggested QbD principles with regular risk assessment sessions early in the drug project, many potential future crystallization problems can be avoided. The most important decision – the choice of solid form- should be based on comprehensive salt- and polymorph screenings, careful API characterization including formulation properties of different solid forms. A complete knowledge about the phase diagram in the chosen process system i.e. understanding the stability limits for the polymorphs, will save a lot of problems. The choice of crystallization procedure is also important. A cooling crystallization in a single solvent or well-defined solvent mixture is more likely to deliver consistent batches than a direct crystallization of a sparingly soluble salt. Anti-solvent crystallizations are very efficient and flexible but often require accurate control of volumes and charging rate of the anti-solvent. This can be difficult when applied immediately after a reaction step.
Phase Diagrams
In many cases, it is possible to isolate the desired form (polymorph, hydrate, salt or co-crystal) of the API under conditions where it is thermodynamically stable. For simple cooling crystallizations containing principally the API and a fixed solvent composition, the thermodynamics can be described completely by a temperature-solubility curve. This can be used to define the screening temperature and isolation temperature, hence the yield and productivity of the cooling crystallization [3]. Because this information is thermodynamic, it is independent of scale – the main requirement on scale-up is that the system achieves equilibrium before filtration commences.
A temperature-solubility curve is like a map. The coordinates of temperature and composition define locations within the map, and the solubility curve separated regions on the map that have different properties, as the coastline separates sea from land. In many API crystallizations, there is more than one solid or more than one solvent. The compositions of such systems cannot be described by one concentration parameter. Examples of systems containing three components include racemic solutions, salt formation, cocrystal formation, anti-solvent additions and hydrate formation from mixed solvents. The ‘maps’ that are most useful in such cases are isothermal, with the axes describing the two ratios necessary to define the composition of the system. These isothermal ternary phase diagrams are familiar to geologists and have been used extensively to describe the behavior of racemic solutions [4, 5]. More recently, they have also been applied to formation of hydrates [6], co-crystals [7] and salts [8]. Here we describe two in-house examples where phase diagrams have been used to define the operating region for a process. Figure 2 shows these phase diagrams, which use the same color coding. The blue regions represent solutions. The red regions represent compositions where two solid phases are both present at equilibrium – these regions are generally to be avoided. In the green and yellow regions there is only one solid present, and in the yellow regions this is the desired solid.
Example 1: Making an Anhydrous Form
In this example, the desired form was anhydrous, but a hydrated form was also known. A basic requirement of the final crystallization was to avoid hydrate formation. The simplest way to do this would have been to exclude water from the crystallization. However, in this case water was essential, to achieve sufficient solubility, and to ensure the removal of inorganic impurities. Moreover, the process included the separation of an aqueous layer followed by distillation, so the amount of water present at the end of the crystallization was different from the amount added initially.
Figure 2, left shows a schematic isothermal ternary phase diagram for such a system. Following the trajectory of the arrow corresponds to starting with a suspension of solid anhydrate in a saturated solution in dry solvent at point A and adding water. Initially the added water resides in the solvent. On entering the red region, the hydrate begins to crystallize. As more water is added, the solution composition stays constant at composition C while the anhydrous form turns over to the hydrate. Once this turnover is complete, we enter the green region in which the solvent composition continues to change as more water is added. In order to be sure that the anhydrous form will be isolated, the water: organic solvent ratio in the system must be less than at point C. This critical water content at the isolation temperature was determined experimentally by a series of slurry experiments using in-line technology to monitor the course of transformations [9].
The stable phase at the end of the crystallization is controlled by keeping the water composition of the system within the range specified by the phase diagram. This could have been verified by in-process tests, but a more elegant solution was to use the batch temperature. Modeling of the distillation process showed that the batch temperature at the end of distillation was sufficiently sensitive to the water content. Ensuring that this temperature was at least 106° C translated to a water level at isolation sufficiently low to avoid the hydrate.
Example 2: Forming a Salt
The solubilities of the acid, base and salt in a fixed solvent composition define which stoichiometries of acid and base will yield the crystalline salt. In this example, the desired product is the 1:1 salt and no other salt stoichiometries are known. The salt is crystallized by mixing solutions of the acid and base at elevated temperature, then seeding and cooling the solution to ensure a low supersaturation process. To avoid impurity formation under acidic conditions, the process is designed with a small excess of base. The system is allowed to reach equilibrium at the final temperature before isolation and so a ternary phase diagram such as that shown in figure 2, right applies.
The key features of this schematic phase diagram arise because the base is less soluble than the acid in the process solvent. Imagine starting from a slurry of the base in the solvent at point F and adding acid to move along the line shown. Initially, the solubility of the base increases, until the system composition enters the red region of the diagram. Now the solution composition is constant at point B, and as further acid is added the amount of solid base decreases as the amount of salt solid increases. Composition B is a eutectic composition and the system is buffered. When the entire base has dissolved, we enter the yellow region which is the correct region for isolation of the pure salt. The line between B and the solid salt separates these regions and defines an edge of failure for the process.
The eutectic point B can be measured by allowing slurries of the salt and base to reach equilibrium at the isolation temperature of the process. The concentration of the process is fixed and so the acid/base composition at the edge of failure can be defined. Slurry experiments showed that at the process composition (represented schematically by X), an acid: base ratio less than 0.89:1 would be required to allow crystalline base to be isolated with the salt. Ensuring an acid charge of greater than 0.89 mole equivalents with respect to the base in the process would therefore avoid contamination of the isolated salt with crystalline base.
Process Analytical Technologies (PAT)
Process analytical technologies have become increasingly important in the pharmaceutical industry in recent years [10]. A diverse range of in-line tools can be used to monitor solution composition and solid form in crystallization processes from laboratory to plant scales. Application of these techniques in process development and scale up experiments is useful in developing process understanding and in many cases this removes the need for in-line monitoring at production scale. PAT can be used to reduce the number of experiments required to develop a process by optimizing the information acquired during experiments. It also allows access to information unobtainable by offline methods.
Solid form can be monitored directly using spectroscopic techniques such as near infrared (NIR) and Raman spectroscopies and indirectly by monitoring changes in the particle size by in-line chord length distribution measurements (in-line CLD measurement). In-line CLD measurement is now a standard technique to monitor crystal nucleation and growth. Short-lived metastable polymorphs or solvates are difficult to capture with off-line XRPD analysis but can be observed with in-line CLD measurement or Raman spectroscopy. Solution composition and supersaturation can be monitored by UV and IR spectroscopy. PAT allow the time of any changes in a process to be captured accurately, avoid changes which may occur during sampling and allow monitoring to be carried out throughout the process. The examples described here illustrate the use of PAT in process development, scale-up and manufacture.
Example 3: Use of In-Line CLD Measurement in Process Development
Early in development a recrystallization process for a basic API was required. The solubility of the API was measured in 10 solvents with diverse properties at 20°C and a solvent with suitable solubility was selected. From this data, a seeded cooling crystallization from isopropanol was expected to be feasible.
A trial process was monitored using in-line CLD measurement and the data are shown in figure 3a. The dissolution temperature was noted and this information was used to define the seed point. The seed were added at 12:30 at 10°C below the dissolution temperature. The total counts increase and remain elevated during the hold at 70°C indicating that the seed crystals do not dissolve. No crystal growth is observed during the hold after seeding but crystal growth occurs during the cooling step and then to a lesser extent during the hold following cooling. Some attrition may also be occurring at this stage. Monitoring by in-line CLD measurement allows us to determine the dissolution temperature and so select a seed point and in the same experiment confirm that adding seed at this temperature is successful. Monitoring during larger scale experiments allows us to determine the effect of scale on the kinetics of the process.
Example 4: Use of In-Line CLD Measurement on Scale-Up
In this case study, the monitoring of one such process on scale up allowed detection of an unknown solid form. The in-line CLD measurement data in figure 3b for a seeded cooling crystallization of an API crystallization at 100l scale showed unexpected dissolution and crystal growth on cooling to 0°C. This occurred after successful seeding and crystal growth of the desired anhydrous form. At 6:00 the crystals of the anhydrous form dissolve and a new solvated crystalline form grows. In the subsequent isolation and drying, the solvated form reverts to the desired anhydrous form. The solvated form would not have been detected without inline monitoring. Undetected, this solvated form would remove the ability to control the particle properties in the crystallization. The use of in-line CLD measurement to monitor the crystallization during development improved process understanding.
Example 5: The Use of In-Line Raman Techniques to Identify Unstable Solvates
Long drying times are sometimes difficult to explain but may be caused by small amounts of solvates that exist as long as the substance is in contact with the solvent. By introducing in-line measurement with Raman spectroscopy during the crystallization of an API a solvate was identified. Raman spectra of the slurry contain information from both the solution and the solid and the solid-state information needs to be extracted by mathematical treatment of the spectra. The characteristic peaks from the solid phase can be obtained by subtracting the corresponding mother liquor spectra as shown in Figure 4. A small, but significant difference over time is then seen around 1150 cm-1. This indicates that the solvate is a metastable form that in the current process only exists in the slurry for a few hours. A rapid transformation to a more stable form occurs and the solvate is not a problem. 
Understanding the Process
Once the solvent and supersaturation technique are decided it is important to investigate the robustness of the process by a systematic variation of the process parameters to capture any kinetic effects. Typical examples are solvent composition, solution concentration, temperatures and crystallization times. It is also necessary to check the influence of the solution purity before crystallization. Repeat the crystallization with very pure material to identify any possible future effects that may occur due to optimized chemistry and improved work-up procedures. Specially designed edge-of-failure experiments will identify combinations of process parameters that must be avoided.
Example 6: Understanding Polymorphism
Example 6 discusses the application of factorial design to identify both safe and unsafe process regions. Substance X exists in two crystal modifications A and B. The wanted polymorphic form (form B) was identified as a critical quality attribute (CQA) and to minimize future CQA failures it was necessary to have a good understanding of the influence of process conditions on the modification. The substance is crystallized in an anti-solvent crystallization at 50 °C, followed by a cooling step down to 0°C. The initial solvent composition (solvent 1 and 2) and the charging rate of the anti-solvent were identified as critical process parameters.
The Use of Factorial Design
Three factors were varied on two levels which gave a 23 set of experiments (full factorial design, without centre points). In addition, two more intermediate levels of addition time (120 and 300) were added. The crystallization was observed and the identity of the isolated product was analyzed with XRPD. The outcome of the factorially designed experiments is illustrated by the parameter cube in Figure 5. It shows that the unwanted Form A (red circles) or a mixture of Form A and B (orange and yellow circles) is obtained at slow charging rates of the anti-solvent (i.e low supersaturation). Form B (green circles) is predominantly obtained at fast charging (high supersaturation levels). Further, the residual solvents in the solution have a strong influence on the crystallization: a high level of solvent 1 and 2 at fast charging rate increases the risk of initial oil formation (squares) before nucleation starts, and also the risk of obtaining form A or mixtures of A and B (yellow circle). The region where Form B is obtained consistently is at low level of solvent 1 and fast addition rates of the anti-solvent (upper left part of the cube).
Conclusions
Control over the final product quality is essential in the pharmaceutical industry. This requires the design of robust and flexible crystallization processes that deliver consistent material with the expected solid state and chemical properties. The use of the quality-by-design approach (Qbd) throughout the life-cycle of a drug product is an efficient tool for faster and more focused process development. It results in greater regulatory flexibility as well as more robust processes. This paper describes three important tools for a successful process design:
- The use of phase diagrams for a complete understanding of the thermodynamics of the system,
- The use of process analytical techniques (PAT) to increase the understanding of the process kinetics,
- The application of factorial experimental design to understand the effect of process variations
Acknowledgement
The authors wish to thank Lena Hedström, Magnus Sjögren, Veronica Profir, Ingvar Ymén, Martin Kenworthy, Oliver Cunningham and Kevin Vare for their assistance in preparing this paper.
References
- ICH Harmonized Tripartite Guideline. Pharmaceutical Development Q8(R1), Step 4 Version, November 2008
- The Gold Sheet, Pharmaceutical &Biotechnology Quality Control, vol 39, no 12 (2005)
- Black, Simon N; Quigley, Kathryn & Parker, Alexander (2006). A Well-Behaved Crystallisation of a Pharmaceutical Compound. Organic Process Research & Development, 10(2). 241-244
- Jacques, J; Collet, A & Wilen, S.H. (1994). Enantiomers. Racemates & Resolutions, Re-issue, Krieger, Florida ISBN 0-89464-876-4.
- Lorenz, L; Sheehan, P & Seidel-Morgenstern (2001). Coupling of simulated moving bed chromatography and fractional crystallisation for efficient enantioseparation. J. Chrom. A, 908, 201-214.
- Jarring, K; Larsson, T; Stensland, B. & Ymen, I. (2006). Thermodynamic Stablity and Crystal Structures for Polymorphs and Solvates of Formeterol Fumarate. J. Pharm. Sci. 95 (5) 1144-1161
- Chiarella, R.A.; Davey, R.J. & Peterson, M.L. (2007). Making Co-Crystals-The Utility of Ternary Phase Diagrams, Crystal Growth & Design, 7 (7) 1223-1226 DOI: 10.1021/cg070218y
- Simon N. Black, Edwin A. Collier, Roger J. Davey, Ron J. Roberts (2007). Structure, solubility, screening and synthesis of molecular salts. J. Pharm. Sci. 96 (5) pp1053-1068, DOI: 10.1002/jps.20927
- Simon N. Black, Andrew Phillips & Claire Scott (2009). Crystallization Design Space: Avoiding a Hydrate in a Water- Based Process. Org. Process Res. Dev. 13(1) 78-83 DOI: 10.1021/op800249r
- Yu, Lawrence X.; Lionberger, Robert A.; Raw, Andre S.; D’Costa, Rosario; Wu, Huiquan; Hussain, Ajaz S. (2004). Applications of process analytical technology to crystallization processes. Advanced Drug Delivery Reviews, 56, 349-369
Martin Bohlin is a Principal Scientist in crystallization at Process R&D Södertälje, Sweden. He received his MSc in Chemical Engineering at the Royal Institute of Technology, Stockholm (RIT), Sweden in 1986 and a PhD in Chemical Engineering in 1993 under supervision from Prof. Åke Rasmuson. Since then, he has been working for AstraZeneca where he has been involved in a large number of drug projects. He has practical and theoretical experience with all types of crystallization development work such as salt- and polymorph screening, resolutions, crystallization method development, scale-up and optimization, supercritical crystallisation, trouble shooting etc.
Simon Black is a Principal Scientist specializing in Crystallization in AstraZeneca’s Process Research and Development Department in Macclesfield, England. He studied Natural Sciences at Cambridge, staying on to complete his Ph.D. in physical chemistry. He joined ICI to work on crystallization at scales from grams to tonnes, moving to AstraZeneca in 2000. Simon has over 40 publications and patents, and is a visiting professor at Manchester and Tianjin Universities.
Helen Jones is a Crystallization Scientist at AstraZeneca PR&D in Macclesfield, UK. She graduated from the University of Bristol in 2002 with an MSci in Chemistry with Study in Continental Europe (Germany) and subsequently researched “Crystallization of Polymorphic Organic Salts” under the supervision of Professor Roger Davey receiving a PhD from the University of Manchester in 2006. She currently works on API crystallisation process development and scale up. Her research interests include speciation in non-aqueous solvents and salt crystallization.