Exploration of the Dynamic Properties of Protein Complexes Predicted from Spatially Constrained Protein-Protein Interaction Networks

Protein complexes are not static, but rather highly dynamic with subunits that undergo 1-dimensional diffusion with respect to each other. Interactions within protein complexes are modulated through regulatory inputs that alter interactions and introduce new components and deplete existing components through exchange. While it is clear that the structure and function of any given protein complex is coupled to its dynamical properties, it remains a challenge to predict the possible conformations that complexes can adopt. Protein-fragment Complementation Assays detect physical interactions between protein pairs constrained to ≤8 nm from each other in living cells. This method has been used to build networks composed of 1000s of pair-wise interactions. Significantly, these networks contain a wealth of dynamic information, as the assay is fully reversible and the proteins are expressed in their natural context. In this study, we describe a method that extracts this valuable information in the form of predicted conformations, allowing the user to explore the conformational landscape, to search for structures that correlate with an activity state, and estimate the abundance of conformations in the living cell. The generator is based on a Markov Chain Monte Carlo simulation that uses the interaction dataset as input and is constrained by the physical resolution of the assay. We applied this method to an 18-member protein complex composed of the seven core proteins of the budding yeast Arp2/3 complex and 11 associated regulators and effector proteins. We generated 20,480 output structures and identified conformational states using principle component analysis. We interrogated the conformation landscape and found evidence of symmetry breaking, a mixture of likely active and inactive conformational states and dynamic exchange of the core protein Arc15 between core and regulatory components. Our method provides a novel tool for prediction and visualization of the hidden dynamics within protein interaction networks.Author Summary: Cells are complex dynamic systems, and a central challenge in modern cell biology is to capture information about interactions between the molecules underlying cellular processes. Proteins rarely act alone; more often they form functional partnerships that can specify the timing and/or location of activity. These partnerships are subject to dynamic changes, and thus protein interactions within complexes undergo continuous transitions. Genetic and biochemical evidence suggest that regulation or depletion of a single protein can alter the stability and activity of an entire protein complex. Experimental approaches that detect interactions within living cells provide critical information for the dynamical system that protein complexes represent; yet complexes are often depicted as static 2-dimensional networks. We have built a system that projects in vivo protein interaction datasets as 3-dimensional virtual protein complexes. By using this method to approximate the diffusion of complex components, we can predict transient conformational states and estimate their abundance in living cells. Our method offers biologists a framework to correlate experimental phenotypes with predicted complex dynamics such as short or long-range effects of a single perturbation to the function of the whole ensemble.