About the CDP Center
The goals of the MIT CDP Center are to develop mathematical models of mammalian signal transduction networks regulating life-death decisions in cells and to test these models experimentally.
It is a truism that detailed knowledge of the parts of a complex system provides only limited insight into the dynamics of the system as a whole. The mathematical emphasis in the CDP is therefore important because only formal models have the power to capture the dynamic behaviors of sets of interacting components. The empirical emphasis is also critical because the validity of network models can only be established by rigorous experimental analysis.
Data-driven mathematical models of biochemical networks controlling cell death and proliferation in human cells are the cornerstones of CDP research. These models range from physico-chemical to statistical. All rely on rich data sets describing changes in protein and mRNA levels, biochemical activities, post-translational modifications, protein localization in live cells etc. as determined from multiple cells and tissues using a variety of techniques. The primary foci of our modeling efforts are pro-death networks downstream of TRAIL, TNF and FAS receptors and pro-survival networks downstream of EGF, IGF and insulin receptors. To date, we have constructed both large-scale statistical models using Dependent Partial Least Squares Regression (PLSR) that describe interactions among receptor families as well as ODE-based mechanistic models that describe the detailed activities of individual receptor pathways. Data collection centers on a pipe-line of commercial and in-house high throughput technologies, as well as significant efforts in the areas of imaging and Microsystems research.
PLSR modeling of crosstalk between death and survival signals in human epithelial cells treated with combinations of TNF, EGF and insulin has revealed that cells respond to TNF directly, via activated TNF receptor and indirectly, via the sequential released of three cytokines: TGF-α, IL-1α, and IL1 receptor antagonist (IL1ra). This contingent and time-varying series of signals constitutes an autocrine cascade linking cellular responses to the local environment and making death cues self-limiting. A practical insight is the close link between drugs that affect pro-inflammatory signaling by TNF and IL1 (e.g. Remicade, Enbrel, Kineret) and those that inhibit ErbB1 (e.g. Iressa and Erbitux).
A potential weakness of principal components analysis (PCA), of which PLSR is a variant, is that information on which dimensions contribute to principal components (PCs) is usually not informative. We were therefore very surprised to learn that our PLSR model was interpretable in this fashion: one PC was composed of activated versions of proteins in the Akt kinase pathway and the other of activated Jnk1, MK2 and caspase 8. When the TNF+insulin and TNF+EGF treatments were plotted against the PCs, the response vectors were nearly orthogonal showing that EGF and insulin oppose TNF-mediated death in very different ways. Surprising, a PLSR model developed for TNF, insulin and EGF was shown to perform well in a completely different biological context: predicting cell death following infection with adenovirus, suggesting that statistical models may be valuable outside the scope of the data on which they were trained.
Recent research in CDP has focused on the construction of ~300 to 500 species ODE models describing particular receptor subsystems involved in the death and proliferation of human cells. These data are trained on quantitative time-course data and validated using RNAi, drugs and other means of perturbing cell behavior. Models of TNF and TRAIL-mediated signaling show the value of single cell measurement in dissecting a snap-action, variable delay switch. We have established that proteins hitherto thought to be redundant for death, such as SMAC and Apaf1, are in fact individually essential for correct switching. We speculate that mutations in apoptotic proteins found in human cancer might lead to abnormal states of partial cell death.
Our most intensive modeling effort has been devoted to ErbB-mediated signaling in tumor cells. Our current model correctly predicts differences in Iressa responsiveness of human non-small-cell lung cancer cells carrying wild-type ErbB1 receptor and ErbB1 with drug-sensitizing mutations. Using a combination of systematic measurement, modeling and reconstitution in transgenic mice we have shown that ErbB1 mutations have at least three effects: increasing Km for ATP, decreasing internalizing rate and changing the spectrum of receptor hetero-oligomers on the cell surface. We believe that further analysis of these effects will influence the design and use anti-ErbB therapeutics. Based on this work we are interesting in extending systems biology concepts to both theoretical and data-driven analysis of pharmacological action. We refer to this emerging area as “computational and systems pharmacology” and envision it as a bridge between genomic medicine and drug discovery.