(72e) Prediction of Gene Expression Using Goal-Directed Metabolic Models

The salient feature of cybernetic models, their goal-oriented control policies, have been used to predict metabolic phenomena ranging from complex substrate uptake patterns, and dynamic metabolic flux distributions to the behavior of gene knockout strains. The key element of this approach is the perspective that metabolic regulation is organized towards achieving goals relevant to an organism’s survival. For example, in multi-substrate cultures of E. coli, cells invest resources into enzymes for pathways which maximize their growth rate. While goal-oriented control has successfully yielded the prediction of numerous metabolic phenomena, the control policy itself has not been compared with data for metabolic regulation. In this talk, cybernetic control variables are weighed against gene expression data, an indicator of cellular regulation, for two sets of systems.

The first set of systems analyzed consists of bacterial cultures that exhibit diauxic growth in two-substrate environments. For these bacterial models, enzymes are synthesized to maximize the growth rate for a given set of metabolic options at each point in time. Both microarray and qRT-PCR data for pathways pertaining to each substrate are compared with the cybernetic variable for enzyme formation. By demonstrating how cybernetic variables for induced enzyme synthesis mimic transcriptional data, a strong argument for using cybernetic models is made.

The second system modeled describes the behavior of a network of lipids in the mammalian macrophage cell line, RAW 264.7. More specifically, arachidonic acid is converted into various eicosanoids associated with inflammation in response to the detection of a key marker of infection, lipopolysaccharide, by Toll-like receptor 4 on macrophages. The existing framework for modeling single-celled organisms has been modified to describe more complex mammalian metabolic networks. Our approach essentially mimics, in an approximate sense, the maximization of the rate of TNF-alpha production, a crucial factor during inflammation, as an objective function. This is implemented by assuming that the upstream generation of eicosanoids concomitantly results in the downstream transcription of TNF-alpha. To be precise, enzyme synthesis for pathways that yield TNF-alpha production signals is regulated in such a way to ultimately maximize the signal for TNF-alpha formation. This is captured in the model using a linear equation that connects time-dependent changes in TNF-alpha with temporal changes in eicosanoid levels. Fitting this model to metabolite data and comparing cybernetic variables with dynamic gene expression data at the major branch point in this network shows a significant level of agreement. This result serves as an important validation of cybernetic variables using real data from cells. Moreover, it shows that cybernetic variables can be used to infer trends in gene expression data only using information taken from the metabolite level and a description of the metabolic network’s organizing principle. This idea has not only been tested for a control condition and LPS treatment but now has also been tested on treatment with compactin, a statin drug that influences arachidonic acid metabolism.