The Dynamics of Protein Interaction Networks
Many proteins achieve their function by interacting with other proteins in the cell. Kinases, for instance, bind their protein substrates in order to transmit information about the extracellular environment, and stable interactions between subunits are crucial for the assembly of molecular machines like the ribosome and the proteasome. Over the past thirty years it has become increasingly clear that, viewed on a cell- or genome-wide scale, protein interactions comprise vast networks whose size and complexity makes it difficult to reason about their behavior.
The primary goal of our research is to understand the dynamics and function of these complex interaction networks. We attack this problem by building mathematical and computational models of specific biological systems, and we test the predictions of these models against experimental data generated by our own lab or collaborators. Current work in my lab is organized around two broad themes: the function of molecular signaling networks and the dynamics of macromolecular assembly.
The evolution of crosstalk in signaling networks
All cellular life relies on signaling networks to sense changes in the environment and mount appropriate responses. In bacteria, for example, those changes might represent fluctuations in the availability of vital nutrients or the presence of other bacterial species. For human cells, environmental signals generally represent growth factors and cytokines secreted as part of paracrine or autocrine signaling. These factors control critical cell fate decisions, such as proliferation and programmed cell death.
Despite their ubiquity, the structure of signaling networks varies in evolution. Bacterial signaling networks are relatively simple, with essentially no crosstalk between cell-surface receptors (panel A, blue) and downstream transcription factors (panel A, red). Human singaling networks, on the other hand, are extremely complex, with high levels of crosstalk between receptors and transcription factors (panel B). It is currently unclear what evolutionary pressures have driven the evolution of such different architectures.
Work from our lab has indicated that this global difference in network topology is likely a consequence of differences in the local interactions that make up the network. In bacteria, cell-surface receptors both activate and deactivate their target transcription factors, depending on the receptor’s activation state. In humans (and other eukaryotes, particularly metazoans), activation and de-activation are carried out by separate proteins (e.g. kinases and phosphatases). Using mathematical and computational models, we showed that bacterial signaling cannot easily tolerate crosstalk, since adding more targets greatly reduces signaling efficiency (Rowland and Deeds PNAS 2014).
Using similar models, we found that the kinase/phosphatase pairs typical of human signaling networks do not share this constraint. Instead, crosstalk tends to make the responses of downstream elements more switch-like, and we demonstrated that those responses can be coupled in interesting and often surprising ways (Rowland, …, Deeds Biophys J 2012). We are currently working to understand how these constraints have influenced the evolution and function of signaling networks in bacteria, humans, and other organisms.
The noise is the signal
Over the past ten years, measurements of molecular responses in individual cells have demonstrated that human signaling networks exhibit high levels of variability or noise. Using information theory, Levchenko and co-workers demonstrated that noise has a massive impact on the ability of signaling networks to transmit information. Their work focused on the channel capacity, which is the maximum amount of information a system can possibly transmit, quantified in bits. They showed that most signaling networks in human cells carry less than 1 bit of information from the environment. This is somewhat surprising, at least prima facie, given the critical nature of many of the cell fate decisions that these networks control.
Using a combination of mathematical models and experimental results from our collaborator Peter Sorger at Harvard Medical School, we have shown that there is an inherent trade-off between the information a cell has about any given signal, and the capacity of that signal to control decisions at the population level (Suderman, …, Deeds PNAS 2017). In cells exposed to pro-apoptotic cytokines (e.g. TRAIL, panel A), apoptosis occurs once the activity of the Initiator Caspase (IC) casp-8 exceeds a threshold set by anti-apoptotic Bcl-2 protein. As with the molecular responses to other signals, we found wide distributions of the activity of both casp-8 and the Effector Caspase (EC) casp-3 in isogenic populations of cells exposed to any particular dose of TRAIL (panel B), resulting in a channel capacity of 0.5 – 1 bits between TRAIL dose and IC or EC response. Interestingly, however, the fraction of cells that die at any given TRAIL dose was highly repeatable, giving rise to a population-level channel capacity of 3 – 4 bits.
Using a series of mathematical models, we found that variability in the responses of individual cells, plus a threshold that leads to a particular cell-fate decision like apoptosis, allows for precise control of cell populations. Indeed, we demonstrated that higher levels of cellular noise can actually compensate for environmental fluctuations of the concentrations of signaling molecules like TRAIL within tissues.
Interestingly, when the key biological unit of response is a single cell (as in chemotaxis or yeast cell mating), we found that single-cell channel capacities at the molecular level are much higher (over 2 bits). Overall, this work suggests that noise in metazoan signaling may be a highly regulated phenomenon; noise is likely suppressed when the response of individual cells is key, and it may be amplified when a multicellular system needs to control a population. We are currently working to better understand how variability within a population might be regulated, and how cellular systems compensate for noise in processes like chemotaxis.
Many of the most important protein complexes in cells, from the proteasome to the apoptosome, consist either of ring-like structures or of rings stacked on top of each other. Although ring-like structures can display incredible thermodynamic stability, their assembly dynamics can be quite complicated. In particular, these complexes can exhibit a form of kinetic trapping that we call deadlock, in which small intermediates are exhausted from the system before 100% of complexes have assembled (Deeds et al. PNAS 2012). We found that hetero-trimeric ring structures have evolved protein interaction interfaces that tend to minimize the impact of deadlock.
Work in our lab is currently working on understanding on how larger, stacked-ring structures avoid deadlock. Our current focus is the proteasome Core Particle (CP), a massive molecular machine that is responsible for the majority of targeted protein degradation in a variety of organisms, including humans, yeast, and actinomycete bacteria like Mycobacterium tuberculosis. We recently found that certain assembly chaperones likely have evolved to prevent deadlock in the assembly of the CP in yeast, by coordinating CP assembly and binding of the CP to other regulatory factors (Wani, …, Deeds and Roelofs Nat Commun 2015). In addition to providing an understanding of how complexes like the proteasome assemble, this work is providing a set of design principles that will be helpful in the development of novel synthetic structures that self-assemble efficiently and robustly in solution.
Another aspect of our work on self-assembly is the development of assembly inhibitors. The proteasome is a major drug target in the treatment of cancer (the proteasome inhibitor Bortezomib is an FDA-approved anti-cancer drug), tuberculosis and other diseases. Since the proteasome is not functional unless it is fully assembled, assembly inhibitors offer an alternative approach to traditional active-site inhibitors. We are using a combination of computational and experimental techniques to discover and develop this new strategy for proteasome inhibition.