Exquisite Particles: Towards predicting agglomeration in APIs

Academic: Prof Alastair Florence, Dr Ian Houson
Researcher: Dr Cameron Brown


The Challenge

Along with nucleation and growth, agglomeration is a commonly occurring process in crystallisation operations. Agglomeration becomes an interesting process to control as, depending on the product requirements, can be both desired and undesired. Undesired agglomeration can have drastic effects on product consistency. The entrainment of mother liquor and impurities between the primary particles of an agglomerate lead to diminished washing efficiency and lower final product purity. Furthermore, fragile agglomerates can break under the stress of filtration, resulting in blockage of filter medium. Whilst it is generally understood what drives agglomeration, quantification of that driving force and prediction of agglomeration behaviour is lacking. This lack of understanding in compounded by the difficulty in being able to regularly quantify the degree of agglomeration occurring.

The Technology

It is very easy for an individual to visually identify an agglomerated particle in comparison to a single crystal. However, manual segregation of every particles into agglomerates and single crystals would be incredibly tedious, time consuming and subject to bias. Whilst not 100% accurate, simple image analysis methods can lead to significant automation of the segregation process. Coupling this with in-situ time lapse images from the Mettler-Toledo PVM probe, trends of particle number, size, shape and transparency can be determined. These trends can then be semi-quantitatively linked to the agglomeration process.

The Outcome

Agglomeration has been shown previously to be driven by a combination of physio-chemical and hydrodynamics processes. For active pharmaceutical ingredients such as paracetamol this is believed to be linked to the hydrogen bonding potential of the various functional groups on the crystal faces.  As a result, the hydrogen bonding ability of the solvent used strongly influences the occurrence of agglomerates. A screening methodology was developed in which seed crystals of known concentration and size where allowed to agglomerate. Over the process in-situ images were recorded and analysed by the developed image analysis algorithms. The result of which was the classification of solvents into 3 classes: rapid agglomerating (< 2 min), agglomerating (> 8 min) and non-agglomerating. Agglomerating solvents were then subject to further tests covering a range of hydrodynamic conditions. Revealing that the solvents which showed rapid agglomeration where unaffected by increasing shear rates. In contrast to the agglomerating solvents demonstrated that the agglomeration process could be influenced by the shear rate, and in some cases prevented entirely.

The Impact

Presentation of results from this project along with training on the image analysis algorithms and general discussions on particle imaging have taken place at AstraZeneca, Macclesfield and GSK, Stevenage (with similar visits to Novartis and Bayer planned). This has led a 3 month industrial placement of Francesca Perciballi at GSK, Stevenage, working on the implementation of the agglomeration screening methodology to an active ingredient.

Although developed initially solely for this project, the image analysis algorithms have grown into a set of programs bundled into the PVMA Toolbox (Particle Vision and Measurement Analysis), which has already seen use in projects at both AstraZeneca and GSK.

The outcome of this project has highlighted the importance of particle surface properties in driving agglomeration. To address this a follow up student project has been started which aims to characterise the surface of a number of seed particles from different production methods and relate these surface properties to the measured rates of agglomeration.


Thanks are given to all members of the Exquisite Particles steering committee for their help and support throughout the project.