Chemometric analysis of fast liquid chromatographic data obtained on monolithic silica columns

Since the introduction of monolithic silica columns on the market, there is an increasing amount of information available about the performance of these columns, their stability and many other column characteristics. At the start of this research project, there were only few publications on this subject and therefore there was need for research in this field of chromatography.
Since monolithic columns allow faster separations than conventional silica columns, we were interested to know whether separations, developed on classical silica columns could be transferred to monolithic ones. One of the aims of this research project was therefore to investigate the transfer of some separation methods to monolithic silica columns of 5 and 10 cm in length. In this project, the transfer of six separation methods was examined. Because of selectivity differences between conventional and monolithic silica columns, three of the six transfers were not successful. However, the selectivity differences observed can also be expected within the population of classical particle-based C18 columns and are not exceptional. The analysis times of the successfully transferred separations were decreased compared to the conventional silica column and by increasing the mobile phase flow rate, the analysis time of the separations could be further reduced, in some cases up to forty times. The transferred and accelerated separations were also found to be robust when some experimental factors are controlled within non-significance intervals. Furthermore, the column ageing was measured during the use of the monoliths. We could conclude that column ageing remained limited.
In the second part of our research project we focused on the use of monolithic silica columns for complex separations. Plant extracts such as the Ginkgo biloba extract, contain many compounds which are difficult to separate, even on conventional silica columns. Because of the low column backpressures obtained on monolithic columns, the separation efficiency of the monoliths can be increased by coupling several monoliths in series, which is not feasible with conventional silica columns. This makes the monoliths interesting in the development of complex fingerprint chromatograms. We developed a fingerprint chromatogram for a standardized Ginkgo biloba extract on two coupled monolithic silica columns of 10 cm in length each. A gradient method consisting of tetrahydrofuran, isopropanol and water resulted in a fingerprint in which 77 peaks are observed within 60 minutes. As detection, both a UV and ELS detector were used, the latter allowing detection of the poor UV absorbing ginkgolides and bilobalide in the extract. The signal-to-noise ratio in the ELS chromatogram was optimized by performing an experimental design to select the optimal operating conditions of the ELSD.
In the third part of this research project, it was investigated whether we could develop fast separation methods for green tea extracts and derive relevant information from the chromatograms to predict the antioxidant capacity of the tea extracts. We showed that using partial least squares regression, a reliable multivariate calibration model can be obtained between the fast chromatograms and the antioxidant capacity of the green tea, the latter measured with a colorimetric method. Prior to model building, the chromatograms were aligned and outliers were removed. With uninformative variable elimination partial least squares, only the relevant chromatographic variables were used for modeling, resulting in an even less complex prediction model. Moreover, it is shown that also incompletely resolved chromatograms, with still shorter analysis times, can be used, resulting in prediction models with similar prediction errors.
In the last part of this research project, we continued working on fingerprint chromatograms, but now of vanilla samples. Not the development of the fingerprints was now focused on, but the pretreatment of the fingerprints prior to chemometric analysis, which includes removal of uninformative baseline, signal compression and alignment. The effect of three different alignment methods on the classification of the vanilla samples by principal component analysis is evaluated. It is found that slightly different classifications are obtained after alignment with correlation optimized warping, semi-parametric time warping and target peak alignment. We also compared the performance of the warping algorithms on several chromatographic data. It was concluded that correlation optimized warping and semi-parametric time warping perform better than parametric time warping. The flexibility of the latter is not high enough to correct complex peak shifts in two directions. Correlation optimized warping performs slightly better than semi-parametric time warping but its disadvantage is the need for optimization of two input parameters, which might be laborious, and its longer computation time. Semi-parametric time warping performs faster and mostly the default input parameters yielded clearly aligned peaks.