Process Characterization for Troubleshooting

Getting to the bottom: A model-based approach for root cause investigations in protein chromatography

Figure: Results of the stoichiometric displacement model (SDM) parameter estimation by curve fitting of calibration runs: Simulations using a mechanistic model (solid lines) are displayed in comparison to the PLS model inline peak deconvolution (dashed lines).

Chromatographic protein purification is subject to various effects which may lead to disturbances in the process. Column aging or process variations can, for example, lead to deviations in the eluate composition and hence to inconsistencies in pool purities and yield. Before carrying these disturbances over into the next batch, their source should be identified and – ideally – eliminated.

Troubleshooting during processing leads to time loss and costs. Causes of deviations are hard to identify visually and manual troubleshooting approaches are both time-consuming and expensive. Especially when considering on continuous processing, automated monitoring and control strategies are therefore attracting more and more interest. In this context, mechanistic models are promising, but have not yet been applied for a process analysis of biopharmaceutical protein purification.

Combination of spectral deconvolution with inverse mechanistic modeling. In the present case study, novel Process Analytical Technology (PAT) for accelerated and automated troubleshooting has been considered. In a first step, a calibration of Partial Least Square (PLS) Regression was performed. By means of this PLS model, selective protein concentration could be calculated from the mid-UV absorption spectra. After importing them into ChromX, the model parameters for all the species were determined by curve fitting.

Fast and automated troubleshooting for deviating chromatograms. Once the model was calibrated, it could serve for trouble shooting further observed chromatograms. In case of deviations in the chromatogram , the critical process parameters were automatically altered in ChromX until the modeled chromatogram fitted the observed one. By means of the outline approach, a valuable modeling tool for root cause investigation was established. Following the procedure, typical disruptive sources, such as the aging of columns or process variations, can be identified rapidly from a simple examination of the chromatogram.

See full paper.

Related Projects