Application of modern reversed-phase peptide retention prediction algorithms to the Houghten and DeGraw dataset: Peptide helicity and its effect on prediction accuracy

Twenty five years ago Houghten and DeGraw published a groundbreaking study of reversed-phase (RP)-HPLC retention of 298 peptide analogs, including 260 peptides coding the positional substitution in a 13-mer molecule with all 20 naturally occurring amino acids [1]. The authors challenged the state-of-the-art assumption that peptide retention can be represented as a sum of individual hydrophobicities of the constituent amino acids, and suggested an additional dependence on the ordering (sequence) of the residues. Here we explore the accuracy of modern peptide retention prediction models when applied to this retention dataset. We find that all of them perform below their claimed prediction accuracies. Clearly, the question raised 25 years ago remains unanswered, despite significant progress in the field over the past few years. Analysis of the prediction errors shows that the vast majority of outliers occur due to the amphipathic character of the framework Ac-YPYDVPDYASLRS-Amide peptide. This indicates that the understanding and quantitative description of stabilization of helical structures upon interaction with C18 phase is underdeveloped and should be a priority moving forward. In this report we also show that the presence of N-cap stabilizing residues increases peptide RP retention and should be taken into account. Capping effects have not been considered in peptide RP-HPLC studies, despite the clear evidence hidden in the quarter-century old Houghten and DeGrawʹs experimental results.