RESUMO
14-3-3s are abundant proteins that regulate essentially all aspects of cell biology, including cell cycle, motility, metabolism, and cell death. 14-3-3s work by docking to phosphorylated Ser/Thr residues on a large network of client proteins and modulating client protein function in a variety of ways. In recent years, aided by improvements in proteomics, the discovery of 14-3-3 client proteins has far outpaced our ability to understand the biological impact of individual 14-3-3 interactions. The rate-limiting step in this process is often the identification of the individual phospho-serines/threonines that mediate 14-3-3 binding, which are difficult to distinguish from other phospho-sites by sequence alone. Furthermore, trial-and-error molecular approaches to identify these phosphorylations are costly and can take months or years to identify even a single 14-3-3 docking site phosphorylation. To help overcome this challenge, we used machine learning to analyze predictive features of 14-3-3 binding sites. We found that accounting for intrinsic protein disorder and the unbiased mass spectrometry identification rate of a given phosphorylation significantly improves the identification of 14-3-3 docking site phosphorylations across the proteome. We incorporated these features, coupled with consensus sequence prediction, into a publicly available web app, called "14-3-3 site-finder". We demonstrate the strength of this approach through its ability to identify 14-3-3 binding sites that do not conform to the loose consensus sequence of 14-3-3 docking phosphorylations, which we validate with 14-3-3 client proteins, including TNK1, CHEK1, MAPK7, and others. In addition, by using this approach, we identify a phosphorylation on A-kinase anchor protein-13 (AKAP13) at Ser2467 that dominantly controls its interaction with 14-3-3.
