Crystallisation can provide a means of tailoring the properties of particles. Colm Campbell, Clarochem Ireland, describes a project in which the process was used to improve the homogeneity of pharma ingredients
Crystallisation and powder technology are seen as important technologies at Clarochem (Ireland), a drug substance manufacturing site of the CFM Group (CoFarmaceutica Milanese). They can be useful for optimising process performance by, for example, improving filtration rates or, perhaps more interestingly, to provide active pharmaceutical ingredients (APIs) with special, tailored characteristics.1-4
In the latter instance, the crystallisation process is tweaked to provide material with a particular shape, size, aspect ratio or some other feature of importance to the secondary process.
This article discusses a case study, where an established crystallisation of a pharmaceutical active, C1, is modified to provide particles that blend homogeneously with another material, C2.
Dry powder, multi-component blending is commonly used in the pharmaceutical industry to provide materials suitable for tabletting or granulation. The most common type used is diffusion blending, which relies on the constant dividing and intermeshing of particles in a suitable blender. Generally, particles of similar size will tend to blend well with each other, although other factors can complicate performance, such as surface roughness, shape, particle density or more subtle issues, such as stickiness, surface charge or surface porosity.5
C1 is a chemical manufactured at Clarochem, while C2 is purchased from a third party. The materials are blended at Clarochem in a bicone blender using a well-established, validated process. Up until 2006, blend failures due to poor homogeneity or sampling were reasonably frequent, triggering time-consuming re-blending and re-testing operations.
Analysis of particle size distributions on a Sympatec Helos-Rodos system showed a variable, multimodal distribution with d90 values >300µm for batches of C1 from the production plant, while C2 showed a consistent, unimodal distribution, with a d90 ~150-160µm. Examples of dry powder particle size distributions for typical C1 and C2 are given in Figures 1 and 2, respectively.
The crystallisation process for C1 involves a straightforward cooling profile, where a solution of the material is cooled from reflux to ~10°C, at a constant agitation rate. The crystals thus produced are isolated in a centrifuge.
It is believed that the occasional blend failures arise from the significantly different particle sizes of the two components. This intuitively obvious hypothesis is supported by the success of a milling operation for C1, which was introduced to alleviate the issue. It typically shaves 50-70µm off the d90 value, which greatly improves blending performance.
This temporary and time-consuming "bottom up" milling process was superseded by a redeveloped crystallisation process, a "top down" solution. A programme to study the crystallisation process was embarked upon, supported by a Mettler Toledo Labmax automated lab reactor (ALR), equipped with a jacketed 2-litre flask and Lasentec FBRM. By understanding the influence of mixing, cooling rate and investigating the possibility of seeding, it was believed a robust process could be developed that would ultimately provide C1 with a consistent, smaller, particle size distribution, likely to provide a better match for C2 in the downstream blending process. Promising processes would be further investigated in the cGMP 30-litre kilo lab, available at Clarochem Ireland.
scaling down
Before investigating the influence of the above-mentioned parameters, an important first step involves successful scale-down of the industrial process: in this case, finding conditions in the laboratory equipment that will provide material with similar particle size distributions to the plant, allowing the establishment of a baseline.
While scaling down process variables likely to affect particle characteristics is straightforward for some variables, such as cooling rate, it is more difficult for mixing, a parameter that can profoundly influence particle size. This arises from the effect of different mixing regimes on crystal growth. Well-mixed systems are likely to disturb crystal nuclei to a great extent, thereby inducing more secondary nucleation events. This ultimately leads to the favouring of nucleation over growth.
Of course, well-mixed systems are more effective at breaking crystals by attrition than poorly mixing counterparts, although this is dependent on crystal morphology.
Agitation rate or even agitator tip speed is a very poor variable for comparing mixing on different scales, as other factors such as agitator type and geometry, vessel geometry, presence or absence of baffling and un-filled head space in the vessel are also of importance.
In the absence of Computational Fluid Dynamics (CFD) modelling, Energy Dissipation (Ei) calculations can provide a reasonable basis for modelling mixing.
Table 1 shows the calculated Ei values for a range of mixing conditions of interest. The shaded rows compare the proposed laboratory conditions with the industrial scale situation. Clearly, use of 720rpm in the Labmax will produce a fairly similar Ei to that of our industrial system, which uses an agitation rate of 56rpm. On this basis, the initial "baseline" experiments employed an agitation rate of 700rpm.
Table 2 details the conditions used for three Labmax runs, as well as information pertinent to the particle size distributions of the isolated products. As these values are similar to those obtained in standard production batches, a successful "scale down" of the process had therefore been achieved and a baseline established.
mixing and cooling rates
Three different agitation speeds - 300, 500 and 700rpm - were used throughout the study and a clear relationship between agitation rate and particle size was demonstrated. Results from three representative runs are presented in (Table 3). As expected, higher agitation rates afford smaller particle sizes.
As the highest agitation rate (and therefore highest Ei) accessible in the Labmax matches the industrial condition, the 30-litre kilo lab system was proposed to provide the next "step change" for Ei. Referring back to Table 1, the proposed kilo lab conditions chosen for accessing this step change were those referred to in the "Fill volume 25%, 531rpm" line, which gives an Ei of ~1.5 W/kg. Experiments on this scale afforded the expected outcome of a further decrease in particle size, where values of ~300µm were recorded.
Cooling rate variation appeared to affect particle size to a lesser extent: Labmax runs carried out with linear cooling profiles between 5-20 hrs appeared to have minimal effect on particle size.
seeding
Inoculation of the crystallisation mixture with seed can profoundly influence the particle and bulk properties of the isolated product cake. Seeding should be carried out at a point reasonably close to the middle of the metastable zone, using the appropriate quality and quantity of seed. Typically, the range used is from 0.1-5wt% (based on the weight of input material), although there are exceptions at both ends of the range. Apart from chemical and polymorphic purity considerations, seed typically needs to be milled, or at least sieved, to provide sufficient active surface area for templating the required morphology.
The metastable zone width (MSZW) was found to be quite wide, ranging from 65-75°C, suggesting that seeding at ~70°C would be effective. This was, indeed, shown to be the case. Some of the initial experiments (Table 4) were run at a low agitation rate, using micronised or coarse seed. As expected, the particle size distribution of the product was strongly dependent on seed size. Under these mixing conditions micronised material was significantly more effective providing product with a d90 close to the desired range.
Interestingly, using the faster agitation conditions, smaller particle sizes were obtainable than under the more poorly mixed regime with coarse seed (Table 5). This focuses our attention on the importance of mixing in the system, even in the presence of seed. These data also illustrate the relative unimportance of cooling rate. As observed with the poorly mixed system, the use of fine seed (in this case, sieved) is also effective in producing small particles in the better-mixed regime (Table 6).
The process was scaled up in the kilo lab, employing the same mixing conditions described for the previous kilo lab scale-up. The desired result was obtained, where the d90 was recorded as 241µm. This material was blended in a lab-scale bicone with standard C2 material and HPLC assay confirmed blend homogeneity. Several lab scale C1 batches, produced with sieved seed, were also use-tested successfully in the lab-scale biconical blender.
Overall, the programme was successful at designing a crystallisation that produces C1 material with a similar particle size distribution to that of C2 material (Figures 1 & 3, respectively). The process will be scaled up and validated in the full-scale plant, using fast agitation and seeding with 1wt% of sieved material at 69±1°C. Other noteworthy points include:
The system has a reasonably large metastable zone width (MSZW), making it suitable for seeding
Output from lab runs involving seeding gives better results than for the non-seeded process using a variety of mixing and cooling conditions
Increased seed loading may give greater benefits, but 1wt% is acceptable in this instance
Mixing is of importance in both seeded and non-seeded regimes, while cooling time is not as critical in this example
Fine (sieved or micronised) seed works best over a range of mixing conditions, while un-treated (coarse) seed is acceptable in well-mixed systems only
Whether the fine seed is micronised or sieved probably does have some influence on the particle size of the product, but both seed types are adequate for our purposes.
Computational Fluid Dynamics would be useful to test the hypothesis that high Ei equates to good mixing in this system.
acknowledgements
Dr Donocadh Lydon performed most of the laboratory work, with support from Dr Declan Maher and Dr Mary Lynch. Joe Dunne was invaluable for the kilo lab scale up and mixing calculations.