Multivariate data analysis of near-infrared spectra of cultivation medium powders for mammalian cells

Author Summary: Nowadays, qualification and control of medium formulations is performed based on simple methods (e.g. pH and osmolality measurement of media solutions), expensive and time-consuming cell culture tests, and quantification of some critical compounds by liquid chromatography. Besides the traditional medium qualification tools, relatively new spectroscopic techniques such as fluorescence spectroscopy, nuclear magnetic resonance, Raman and NIR spectroscopies or a combination of these techniques are increasingly being applied for cultivation medium powder investigation. A chemically defined cultivation medium powder for Chinese hamster ovary (CHO) cell cultivation was investigated in this study, regarding its response to heat treatments with different temperatures (30n°C, 50n°C and 70n°C). The heat treatments were performed according to a design of experiments (DoE) approach. Spectra of the control and the treated powders were collected to compare the sample groups using a dispersive near-infrared (NIR) and a Fourier-transform near-infrared (FT-NIR) spectrometer. Multivariate data analysis including unsupervised (cluster analysis, principal component analysis, polar qualification system) and supervised classification methods (linear discriminant analysis, soft independent modelling of class analogies and partial least squares discriminant analysis) were employed to investigate the monitoring capability of near-infrared spectroscopy for qualification of cultivation medium powder stored at different temperatures, to identify the treatment-induced variations in the samples, and compare the efficiency of spectrometers with distinct optical arrangements (i.e. dispersive and Fourier transformed spectrometers). During heat treatment, cultivation medium powders went through spectroscopically recognisable changes. Both NIR and FT-NIR analysis could separate samples according to the temperature set-points, irrespective of spectrometer attributes. In classification of samples cluster analysis and linear discriminant analysis shown the best results for both NIR and FT-NIR spectra