Integrating Expression Profiles into a Constraint Based Model to Infer the Carbon Source of a Micro-Organism

Introduction: In biomedical research and in the public greater focus on pathogenic microorganisms is being taken as their resistance to antibiotics increases and it is debated that it is only a matter of time until the ticking time bomb explodes and multi resistance strains are getting exponentially causative for deadly infections. Hence, specific therapy is needed and, in line, investigation on the specific needs and features of a pathogen within its host environment. Essential to this are the nutritional needs of a pathogen. Besides this, generating transcription profiles on a large scale is technically straightforward and feasible for any kind of clinical isolates of nearly all conditions.

Objective: We aimed to develop a means to predict the nutritional sources of microorganisms from gene expression profiling data.

Method: We have set up a computational prototype to infer the main carbon source of a microorganism based on expression profiles. We focused our investigations to central energy metabolism, generated a consistent constraint-based model of central energy metabolism starting from a generic model, optimized the difference of flux gradients and differential expression of several experimental conditions using linear programming and corrected for futile cycles by penalizing reactions with negative epsilons during optimization. We integrated binary switches into the linear models leading to a Mixed Integer Linear Programming problem. This enabled the optimized selection of the relevant expression profiles being consistent with the flux constraints. For each condition, the transporters with the highest predicted fluxes were determined. As a case study, we applied our method to expression data of Bacillus subtilis 168 in eight different nutritional conditions (in triplicate) (Nicolas, et al., 2012), used and adapted a published model (Oh, et al., 2007) and compared our model predictions to well derived fluxes from a detailed C13 tracing analysis (Chubukov, et al., 2013).

Results: In six out of eight cases (glucose, fructose, gluconate, glycerol, malate, succinate+glutamate) our predictions were correct. Besides this, the predicted fluxes from our model were well comparable to the fluxes from the gold standard (Pearson correlation r=0.58). In contrast to the previous study by Chubokov et al., we found in particular the expression profiles of the transporters as a necessary backbone for constructing realistic expression based flux models.

Conclusion: Our new approach is promising and led to good predictions for this case study of B. subitilis. In particular, the implementation of binary switches and expression gradients of the transporters are very helpful to reconstruct the metabolic fluxes. Our approach may also be used for identifying the carbon sources of isolates from clinically complex chronical settings like e.g. chronically S. pneumonia or S. aureus infected lung epithelial or osteoblasts for which C13 labeling experiments are not feasible.

References

Chubukov, V., et al. (2013) Transcriptional regulation is insufficient to explain substrate-induced flux changes in Bacillus subtilis, Molecular systems biology, 9, 709.

Nicolas, P., et al. (2012) Condition-dependent transcriptome reveals high-level regulatory architecture in Bacillus subtilis, Science, 335, 1103-1106.

Oh, Y.K., et al. (2007) Genome-scale reconstruction of metabolic network in Bacillus subtilis based on high-throughput phenotyping and gene essentiality data, The Journal of biological chemistry, 282, 28791-28799.