Chromatographic models to predict the elution of ionizable analytes by organic modifier gradient in reversed phase liquid chromatography

The retention of an ionizable analyte under RP-HPLC organic modifier gradient elution is strongly affected by its ionization degree which, in turn, depends on its pKa and on the pH of the mobile phase. The values of both parameters change depending on the mobile phase composition and thus retention becomes a parameter quite difficult to predict, particularly when working in gradient mode. In this work, an equation describing the retention of ionizable analytes has been combined with three different models of different complexity, developed for gradient elution of neutral compounds (1, 2, or 3 fitting parameters), in order to predict retention of compounds with acid–base properties with particular buffers. All models have been tested under 16 different gradient patterns (4 linear gradients, 4 concave gradients, 4 convex gradients and 4 combinations between them) for the prediction of the retention time of 12 acid–base compounds (pKa values from 4 to 9) in 3 different buffered mobile phases (pH 5, pH 7 and pH 9) with acetonitrile as organic modifier. The agreement between the experimental and calculated retention times is good for all models. The best results are obtained through the model that depends on three parameters and the accuracy of the two-parameter model is slightly lower but very acceptable too. On the other hand, the predictions performed with the one-parameter model are the less accurate, but good enough to become a valid model taking into account that it requires very little experimental work.

► Three different models to predict the retention of ionizable analytes involving three, two and one adjustable parameters, respectively are proposed.
► The proposed models are successfully validated for different buffered mobile phases and gradient patterns.
► The accuracy of the obtained results for the three- and two-parameter models is good and equivalent.
► The accuracy of the obtained results for the one-parameter model is less accurate but good enough taking into account that very little experimental effort is required.