“We were working on a large scale to model as many complexes as possible,” says Aloy. Proteins that purified together were assigned to functional groups. The authors built three-dimensional models for as many proteins in these groups as possible, based on known structures and protein homologies. They then predicted which proteins interact directly, and subsequently modeled the structures of complexes containing multiple proteins.
Additional structural data came from EM analyses of the complexes that purified with sufficient quality. “We figured out not just who interacts with whom, but how,” says Aloy. “Understanding function requires structure. At the end of the day it's what gives you the biochemistry.”
Using known two-hybrid interactions and estimates based on homology, the group also predicted communications between complexes. Some were unexpected connections, such as those between transcription and translation components. Although the accuracy of many of their cross-talk predictions is unknown, the structures suggest suitable sites for mutagenesis by any group interested in a particular interaction pair.
So far, the authors have a good idea of the structure of about a quarter of the estimated total protein complexes in yeast (∼400) and has nearly complete structures for 42 complexes. “Our final goal,” says Aloy, “is to model all the associations of all the complexes or organelles at a molecular level.” More structural information should be forthcoming once the group is able to improve their EM using tomographic techniques. ▪