Pate, Martin Eric; (1999) A practical investigation into the use of principal component analysis for the modelling and scale-up of high performance liquid chromatography. Doctoral thesis (Ph.D), UCL (University College London).

Text out.pdf Download (13MB)

Abstract

Liquid chromatography is becoming increasingly important for the final purification of biomolecules. Traditionally, chromatography has been modelled using mathematical techniques which require experimental determination of the physicochemical data for the separation of interest. These methods are both time-consuming and very complex and have only been truly successful in the prediction of binary separations and where the species in a mixture do not interact significantly. This thesis investigates the use of Principal Component Analysis (PCA), a multivariate statistical technique for the modelling and scale-up of chromatographic separations using a range of stationary phase chemistries and explores the utility of the approach as an engineering tool for rapid process development. The reversed-phase separation of a semi-purified erythromycin feed was used as the feedstock throughout the study. Separations were performed on columns with various geometries and stationary phases. An isocratic solvent comprising 45:55 (v/v) of acetonitrile/water was used to effect the separation into 4 major components and at least 10 minor components. In a first set of chromatograms, experimental design techniques were used to investigate the effects of four process variables (load volume, load concentration, temperature and pH of buffer) on the chromatogram shapes from each of 4 columns. The choice of appropriate data pre-processing prior to PCA was investigated in order to achieve maximal analytical performance from the statistical method. The results showed that when the retention times of elution peaks changed due to variations in temperature and pH, it was necessary to align the main product peak in order to gain most benefit from the PCA. Correlations were derived which enabled accurate chromatogram predictions (>95%) to be made using data from columns with fivefold change of scale and a fivefold change in sample size (25-fold scaling factor overall). A second set of chromatograms were generated in which only the amount of sample load was varied. Sample concentrations of 20mg/mL were separated by each column, the sample volume applied being in the range 1-10% of the bed volume of each column which included realistic non-linear, overload conditions. The principal components derived reflected similar properties of the chromatograms regardless of scale and stationary phase. These similarities were correlated, enabling predictions to be made from small (5cm length x 4.6mm diameter) to large-scale (up to 60mm diameter). The overall scaling factor in this set of chromatograms was in the region of 5000-fold. The use of data reduction techniques was investigated throughout the study so as to minimise the number of runs required at large-scale whilst maintaining highly accurate (>98% accuracy) predictions. Results showed that considerable reductions in the sizes of data sets could be made (>75% reduction) without significant loss in the quality of the data provided that attempts were not made to extrapolate too far outside the limits of the experimental conditions. PCA appears to be a very promising technique for the rapid and reliable modelling and scale-up of performance requiring minimal experimental work and achieving greater accuracy than with traditional mathematical approaches and makes recommendations for future work involving examining the potential relationships between PCA models and the underlying physico-chemical events controlling chromatographic separations.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item