Integrating protein copy numbers with interaction networks to quantify stoichiometry in mammalian endocytosis
Proteins that drive processes like clathrin-mediated endocytosis (CME) are expressed at various copy numbers within a cell, from hundreds (e.g. auxilin) to millions (e.g. clathrin). Between cell types with identical genomes, copy numbers further vary significantly both in absolute and relative abundance. These variations contain essential information about each proteins function, but how significant are these variations and how can they be quantified to infer useful functional behavior? Here, we address this by quantifying the stoichiometry of proteins involved in the CME network. We find robust trends across three cell types in proteins that are sub- vs super-stoichiometric in terms of protein function, network topology (e.g. hubs), and abundance. To perform this analysis, we first constructed the interface resolved network of 82 proteins involved in CME in mammals, plus lipid and cargo binding partners, totaling over 600 specific binding interactions. Our model solves for stoichiometric balance by optimizing each copy of a protein interface to match up to its partner interfaces, keeping the optimized copies as close as possible to observed copies. We find highly expressed, structure-forming proteins such as actin and clathrin do tend to be super-stoichiometric, or in excess of their partners, but they are not the most extreme cases. We test sensitivity of network stoichiometry to protein removal and find that hub proteins tend to be less sensitive to removal of any single partner, thus acting as buffers that compensate dosage changes. As expected, tightly coupled protein pairs (e.g. CAPZA2 and CAPZB) are strongly correlated. Unexpectedly, removal of functionally similar cargo adaptor proteins produces widely variable levels of disruption to the network stoichiometry. Our results predict that knockdown of the adaptor protein DAB2 will globally impact the stoichiometry of most other cargo adaptor proteins in Hela cells, with significantly less impact in fibroblast cells. This inexpensive analysis can be applied to any protein network, synthesizing disparate sources of biological data into a relatively simple and intuitive model of binding stoichiometry that can aid in dynamical modeling and experimental design.