Global retention model based on multisample system parameters for optimising chromatographic fingerprints of medicinal plants.

Abstract

Listen

Translate article

Developing analytical methods for Traditional Medicine products by liquid chromatography is challenging due to their chemical complexity and the lack of analytical standards for numerous, unidentified constituents. Regulatory agencies recommend chromatographic fingerprint analysis for quality evaluation, relying on peak detection to ensure resolution. Conventional modelling struggles to optimise experimental conditions for such complex samples. Recent research highlights that global models, which integrate column/solvent, and solute-specific parameters, effectively streamline method development. This work demonstrates the feasibility of characterising separation systems using chromatograms from independent experimental designs across multiple plants (lemon balm, green tea, and linden), which are processed altogether to build global models. The multisample system parameters enable the optimisation of gradient programs, not only for the three plants used to develop the generalised global model, but also for external plants (peppermint and pennyroyal), with the acquisition of only an additional single gradient experiment. The MATLAB Global Optimisation Toolbox is tested and compared with the previously applied alternating regression method, providing a convenient and accessible solution for fitting global models. The agreement between experimental and predicted optimal separations found with global models is comparable to the accuracy achieved with models specifically tailored to each solute. This research demonstrates the benefits of enhancing global models with data from multiple plant species, enabling robust characterisation of column and solvent properties. The generalised models allow transferable system parameters and accurate predictions for new plants with minimal experiments. This approach effectively separates solute retention effects from column and solvent influences, achieving predictive capabilities comparable to traditional individual retention models, without requiring standards for each compound.